Anacleto Cortez e Correia Mestre Quality of Process Modeling Using BPMN: A Model-Driven Approach Dissertação para obtenção do Grau de Doutor em Engenharia Informática Orientador : Professor Doutor Fernando Manuel Pereira da Costa Brito e Abreu, Professor Associado, DCTI/ISCTE-IUL e CITI/FCT/UNL Co-orientador : Professor Doutor Vasco Miguel Moreira do Ama- ral, Professor Auxiliar, CITI/FCT/UNL Júri: Presidente: Luís M. M. da Costa Caires, Professor Catedrático, DI/FCT/UNL Arguentes: Geert Poels, Professor Catedrático, Ghent University, Bélgica Toacy Cavalcante de Oliveira, Professor Auxiliar, UFRJ Vogais: Ana Maria Dinis Moreira, Professora Associada, DI/FCT/UNL Fernando Brito e Abreu, Professor Associado, DCTI/ISCTE-IUL Paulo Rupino da Cunha, Professor Auxiliar, DEI/FCT/UC Miguel L. B. Mira da Silva, Professor Auxiliar, DEI/IST/UTL January, 2014
Transcript
Anacleto Cortez e Correia
Mestre
Quality of Process Modeling Using BPMN:A Model-Driven Approach
Dissertação para obtenção do Grau de Doutor emEngenharia Informática
Orientador : Professor Doutor Fernando Manuel Pereira daCosta Brito e Abreu, Professor Associado,DCTI/ISCTE-IUL e CITI/FCT/UNL
Co-orientador : Professor Doutor Vasco Miguel Moreira do Ama-ral, Professor Auxiliar, CITI/FCT/UNL
Júri:
Presidente: Luís M. M. da Costa Caires, Professor Catedrático, DI/FCT/UNL
Arguentes: Geert Poels, Professor Catedrático, Ghent University, BélgicaToacy Cavalcante de Oliveira, Professor Auxiliar, UFRJ
Vogais: Ana Maria Dinis Moreira, Professora Associada, DI/FCT/UNLFernando Brito e Abreu, Professor Associado, DCTI/ISCTE-IULPaulo Rupino da Cunha, Professor Auxiliar, DEI/FCT/UCMiguel L. B. Mira da Silva, Professor Auxiliar, DEI/IST/UTL
January, 2014
[This page is intentionally blank]
iii
Quality of Process Modeling Using BPMN:A Model-Driven Approach
A elaboração desta tese beneficiou do regime de isenção de propinas de doutoramento, no âmbitodo Protocolo de Cooperação existente entre a faculdade de Ciências e Tecnologia da UniversidadeNova de Lisboa e o Instituto Politécnico de Setúbal.
A Faculdade de Ciências e Tecnologia e a Universidade Nova de Lisboa têm o direito,perpétuo e sem limites geográficos, de arquivar e publicar esta dissertação através de ex-emplares impressos reproduzidos em papel ou de forma digital, ou por qualquer outromeio conhecido ou que venha a ser inventado, e de a divulgar através de repositórioscientíficos e de admitir a sua cópia e distribuição com objectivos educacionais ou de in-vestigação, não comerciais, desde que seja dado crédito ao autor e editor.
[This page is intentionally blank]
To Flora, Alexandre and Guilhermeand to the memory of my mother
[This page is intentionally blank]
Acknowledgements
I would like to thank all the people who have helped me through the years along theroute.
I thank my supervisors, Professor Doutor Fernando Brito e Abreu and ProfessorDoutor Vasco Amaral, as well as the members of the CAT, Professors Mira da Silva andPaulo Rupino for helping me through their comments and reviews in several phases ofthis work, namely when it was important to focus on the core of the research and delim-iting its scope.
I would also like to acknowledge the UAIIDE-IPS, specially Sandra Silva and Cláu-dia Louro, for providing me all the necessary logistic support. I want to emphasize myspecial appreciation to the President of the Institituto Politécnico de Setúbal (IPS), ProfessorArmando Pires, for his strategic vision and the contribution for the enhancement of theIPS’s prominence as academic institution. The leadership of Professor Armando Piresallowed IPS’s lecturers and myself, to benefit from the Programa Formação Avançada hepromoted. This praiseworthy initiative was subsequently supplemented by the PROTECprogram sponsored through the Fundação para a Ciência e Tecnologia.
I express my gratitude to the teachers and colleagues of the Departamento de Infor-mática of the Faculdade de Ciências e Tecnologia, Universidade Nova de Lisboa (FCT/UNL),namely those whose classes I attended in the preparation’s year of the PhD, ProfessorsJosé Alferes, João Araújo, Miguel Monteiro, and Miguel Goulão.
A special acknowledgment goes to my colleague Jorge Freitas, who helped me uponthe academic integration at the FCT. My PhD’s compagnons de route Sérgio Bryton andLuís Silva, also helped me with their comments and support. I also greatly acknowledgemy fellows José Pereira and Nuno Gonçalves at Escola Superior de Tecnologia (IPS-EST),who assisted me collecting data for the quasi-experiment carried out on this dissertation.
To all who made bearable the challenging moments through their friendship, spe-cially Francisco Mendeiros, Nelson, Judite, Martiniano, and their families.
My special thanks to my wife Flora, and my sons Alexandre and Guilherme, for theirlove and support.
vii
[This page is intentionally blank]
Abstract
Context: The BPMN 2.0 specification contains the rules regarding the correct usage ofthe language’s constructs. Practitioners have also proposed best-practices for producingbetter BPMN models. However, those rules are expressed in natural language, yieldingsometimes ambiguous interpretation, and therefore, flaws in produced BPMN models.Objective: Ensuring the correctness of BPMN models is critical for the automation ofprocesses. Hence, errors in the BPMN models specification should be detected andcorrected at design time, since faults detected at latter stages of processes’ developmentcan be more costly and hard to correct. So, we need to assess the quality of BPMNmodels in a rigorous and systematic way.Method: We follow a model-driven approach for formalization and empirical valida-tion of BPMN well-formedness rules and BPMN measures for enhancing the quality ofBPMN models.Results: The rule mining of BPMN specification, as well as recently published BPMNworks, allowed the gathering of more than a hundred of BPMN well-formedness andbest-practices rules. Furthermore, we derived a set of BPMN measures aiming to pro-vide information to process modelers regarding the correctness of BPMN models. BothBPMN rules, as well as BPMN measures were empirically validated through samples ofBPMN models.Limitations: This work does not cover control-flow formal properties in BPMN models,since they were extensively discussed in other process modeling research works.Conclusion: We intend to contribute for improving BPMN modeling tools, through theformalization of well-formedness rules and BPMN measures to be incorporated in thosetools, in order to enhance the quality of process modeling outcomes.
ix
x
Keywords: business process modeling - measurement - measures - model driven engi-neering - model driven architecture - BPMN - quality
Resumo
Contexto: O standard BPMN 2.0 contém regras sobre a correta utilização dos elementosda linguagem. Peritos da indústria têm também proposto boas práticas para o melhordesenho de modelos BPMN. No entanto, essas regras estão expressas em linguagem na-tural, sendo por isso susceptíveis de gerar múltiplas interpretações e consequentementemodelos incorrectos.Objectivo: Garantir a correcção dos modelos BPMN é fundamental para a automaçãode processos. Erros existentes em modelos BPMN, devem ser detectados e corrigidosdurante o seu desenho, e não em fases posteriores do desenvolvimento dos processos,onde se torna mais caro e difícil efectuar essa correção. É necessário por isso avaliar aqualidade de modelos BPMN de uma forma rigorosa e sistemática.Método: Foi seguida uma abordagem orientada por modelos para a formalização evalidação empírica de regras de construção de modelos BPMN, bem como de métricaspara modelos BPMN, a fim de melhorar a qualidade dos modelos produzidos.Resultados: A recolha de regras de embebidas na especificação BPMN, bem como emobras recentemente publicadas sobre BPMN, permitiu coligir mais de uma centena deregras de construção e de boas práticas no desenho de modelos BPMN. Além disso,foi proposto um conjunto de métricas BPMN com o objetivo de fornecer informaçõesrelacionada com a correcção dos modelos BPMN, durante a modelação de processos.Tanto as regras como as métricas BPMN, foram empiricamente validados através deamostras estatísticas.Limitações: Este trabalho não cobre propriedades formais relativas ao controlo de fluxode modelos BPMN.Conclusões: O presente trabalho pretende contribuir para a melhoria de ferramentasde modelação BPMN, embebendo nelas regras de construção de modelos, assim comométricas, a fim de melhorar os resultados da modelação de processos BPMN.
xi
xii
Palavras-chave: modelação de processos de negócio - medição - medidas - engenhariaorientada por modelos - arquitectura orientada por modelos - BPMN - qualidade
B A Catalog of BPMN Patterns and Anti-Patterns (Sample) 271
B.1 A Top-Level Process can only be instantiated by a restricted set of StartEvents types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271
B.2 Outgoing Sequence Flow not allowed in an End Event. . . . . . . . . . . . 272
B.3 Outgoing Message Flow not allowed in a Catch Event. . . . . . . . . . . . 273
B.4 A Catch Event with incoming Message Flow must have Message or Multi-ple type . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274
B.5 Explicit Start/End Events do not allow Activities or Gateways without in-coming/outgoing Sequence Flow . . . . . . . . . . . . . . . . . . . . . . . . 275
B.6 A conditional Sequence Flow cannot be used if there is only one sequenceflow out of the element . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 276
B.7 A Boundary Event must have exactly one outgoing Sequence Flow (unlessit has the Compensation type) . . . . . . . . . . . . . . . . . . . . . . . . . . 277
B.8 Use a Timer Intermediate Event with an Event Gateway . . . . . . . . . . . 278
B.9 Use a Default Condition at an Exclusive Gateway . . . . . . . . . . . . . . 279
B.10 Two Activities in the same Process should not have the same name . . . . 280
C Data Collection for a Survey on Effectiveness of Current BPMN Tools on De-tection of Rules Violations on Process Models 281
D Sample of BPMN Process Models used in Empirical Validation 287
E Well-formedness Rules for BPMN Metamodel 295
F A Business Process of Financial Services Provisioning 301
7.9 The USE environment loaded with BPMN metamodel and the BPMN dia-gram presented in Figure 7.10 . . . . . . . . . . . . . . . . . . . . . . . . . . 139
7.10 A BPMN simple diagram of a transactional sub-process . . . . . . . . . . . 1397.11 Data Collection of BPMN process models for empirical validation . . . . . 142
8.1 The BPMN empirical study framework . . . . . . . . . . . . . . . . . . . . 1508.2 Pareto diagrams for BPMN elements and rules . . . . . . . . . . . . . . . . 1678.3 Radar diagram depicting the BPMN elements’ usage by Source . . . . . . . 168
9.1 Radar diagram depicting the BPMN measures by Source . . . . . . . . . . 210
B.1 A Top-Level Process can only be instantiated by a restricted set of StartEvents types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
B.2 Outgoing Sequence Flow not allowed in an End Event . . . . . . . . . . . 273B.3 Outgoing Message Flow not allowed in a Catch Event . . . . . . . . . . . . 273B.4 A Catch Event with incoming Message Flow must have Message or Multi-
coming/outgoing Sequence Flow . . . . . . . . . . . . . . . . . . . . . . . . 275B.6 A conditional Sequence Flow cannot be used if there is only one sequence
flow out of the element . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 276
LIST OF FIGURES xxi
B.7 A Boundary Event must have exactly one outgoing Sequence Flow (unlessit has the Compensation type) . . . . . . . . . . . . . . . . . . . . . . . . . . 277
B.8 Use a Timer Intermediate Event with an Event Gateway . . . . . . . . . . . 278B.9 Always use a Default Condition with an Exclusive Gateway . . . . . . . . 279B.10 Activities on the same Process should have different names . . . . . . . . . 280
Context: The quality of process models, as outcome of the process modeling activity iscrucial, since these models are used in later phases of the process life cycle.Objective: To establish the main drivers of the research to be done on process modelingquality.Method: The elicitation of the research problems, research questions and thesis state-ment.Results: The main contributions expected from this dissertation were elicited, namely(1) the formalization of a set of well-formedness rules that could enforce the BPMNprocess models quality; and (2) the derivation of a set of measures for assessing theinternal and external attributes of BPMN process models.Limitations: The goodness of contributions must be assessed by empirical studies.Conclusion: A road-map outlines and delimits the research work to be done in thisdissertation.
3
1. INTRODUCTION 1.1. Introduction
1.1 Introduction
Processes are a set of coordinated activities and procedures fueled by resources, aimingto fulfill specific goals of particular stakeholders [MaP01]. In this dissertation, we areconcerned with the modeling of specific type of processes: blended set of automated andmanual activities that take place in organizations (see sections 2.2 and 2.3).
A process can be formally documented, or merely informally defined. Regardlessits nature, processes exhibit features that make them suitable for usage of modeling tech-niques. Business process modeling is the set of activities conducted for visually depictingqualitatively grounded models of organization’s processes, so that processes can be ana-lyzed, monitored and improved regarding their expected value.
The design of processes through process modeling is the activity carried out in theearly stages of a process elicitation. This activity helps to identify problems in the be-ginning of the process development [RCG+09] and assists the design of valid processmodels. Moreover, a suitable analysis and verification of processes at the modeling stagewould make easier the process maintenance tasks [ARGP06] by reducing its implicitcosts. This is why efforts should be made to impose quality characteristics to the processmodels.
Following a taxonomy of Mylopoulos et al. regarding quality attributes in softwaresystems [MCN92], in this dissertation we choose the product-oriented approach to deal withthe formal treatment of quality in process modeling (detailed in section 2.4). The product-oriented approach attempts to define methods and tools to ensure that process modelingoutcomes, i.e the process diagrams, can be evaluated to the degree to which they meetcertain external quality attributes (e.g. correctness, understandability, maintainability).
For representing quality characteristics of the process models we chose Business Pro-cess Model and Notation (BPMN) [BPM11] since it is nowadays the most well equippedprocess modeling language [RIRG05, RRIG09] (see section 2.5). BPMN is backed up byOMG, and is currently the business process notation most used among process model-ing practitioners [HW11]. BPMN language definition is based upon a metamodel and isone of the languages with more modeling tools available1 (section 2.5.3). However, sinceBPMN rules are informally expressed in natural language there are weaknesses (see de-tails in section 3.4) identified in the BPMN standard that hinders the production of goodquality process models using currently available tools.
To evaluate the quality of process models, as with the software artifacts, we are awarethat quality is a multi-dimensional concept, with several characteristics. The ISO/IEC25000 series (SQuaRE) [ISO05b], is a well known example where the factors contributingto quality are grouped to constitute the basis of a quality model. A quality model oftendecomposes the view of quality into different levels of quality characteristics. The inter-relationships between quality characteristics and their measures are also documented.Eventually, the internal attributes are linked to external quality attributes. In the case
1http://www.bpmn.org/ [accessed in Feb. 10, 2013]
4
1. INTRODUCTION 1.2. Research Drivers
of process modeling, this results in a quality model with directly measurable attributesof process models diagrams (e.g. size, complexity) linked to the perceived quality bystakeholders (e.g. correctness, understandability, maintainability). So, since a qualitymodel is a framework for quantifying and linking different quality characteristics, onecould expect that if we are able to control the internal quality characteristics of processmodels, this will ensure a more control over the final quality of the modeling process.
1.2 Research Drivers
The starting point of our research process [WRH+00] was to formulate the main prob-lems we were trying to tackle (section 1.2.1). This drove us to the research questions(section 1.2.2). To answer these questions we have to formulate some hypotheses. Theexplanatory theory of the problems posed by the research questions is part of the researchhypotheses formulated in sections 8.4.1.2 and 9.4.2.
In chapters 5 and 6 are described the research work we have performed, from whichwe distilled the contributions referred in section 1.2.4.
For validation of our contributions, we set up some empirical studies (described inchapters 8 and 9). Based on the conducted empirical studies, we drew conclusionsabout the benefits and effectiveness of our proposals.
1.2.1 Research Problems
The aim of our research is to face quality issues on process models raised by the usageof the BPMN 2.0 standard. Those issues can be tackled facing the two following majorsub-problems2:
1. [RPA] BPMN is intended for modelers with different levels of expertise and tech-nical backgrounds, namely process analysts and process developers. However, theBPMN standard does not provide a rigorous definition that modelers could followto produce good quality models. There is a lack on BPMN rules formalizationin the BPMN standard’s documentation. Furthermore, due to the informal BPMNspecification, amenable to originate different interpretations, BPMN tools do notprovide an accurate and in-depth verification of process models (see details in sec-tion 3.4).This problem is important in the specific case of BPMN, given the number andtype of constructs available in the language (see Figure 3.2 and 3.3). Furthermore,if models become large and complex, or several variants for the same problemare considered, checking rules compliance becomes fundamental and difficult toachieve without automatic tools. Therefore, well defined and rigorously formal-ized rules are required for obtaining good quality models in BPMN.
2We label the problems as RPA and RPB for the sake of traceability with research questions, and researchhypotheses
5
1. INTRODUCTION 1.2. Research Drivers
2. [RPB] The BPMN standard does not provide guidelines or measures regarding theinternal and external attributes that well designed process models must possess.According a survey we made on BPMN tools (section 3.4), most of them do not im-plement sound measures, validated by empirical results that could support processmodelers when designing BPMN diagrams.
1.2.2 Research Questions
With the elicitation of the actual problems, in the previous section, we can now comeup with the following research questions, each of one related with the previous researchproblems, which will direct our research.
[RQA] Could the formalization of BPMN rules contribute for uncovering quality nonconformance, as well as attaining better quality of BPMN models? (see section8.3.2)
[RQB] What measures could assess internal and external characteristics of BPMN mod-els? (see section 9.3.2)
1.2.3 Thesis Statement
Grounded in the research problems (section 1.2.1), the research questions (section 1.2.2)and the model-driven approach (section 7.2), we summarize the purpose of this disser-tation in the following thesis statement:
In order to improve BPMN models’ quality we propose a model-driven approachcapable of formalizing: (i) well-formedness rules; (ii) measures for assessment ofmodels characteristics.
Since quality is a multidimensional concept (section 2.4), in the present dissertation,for the sake of focus, we are only concentrated upon the quality characteristics of cor-rectness of BPMN process models. We will further research (see section 10.3) whetherthe approach pursuit in this research work, can also be applied, mutatis mutandis to otherdimensions of quality (e.g. understandability, maintainability).
1.2.4 Main Contributions
The main contribution intended by this dissertation regarding process modeling withBPMN is relate with the improving of the quality of process models at design phase(build-time). This general contribution can be decomposed on the two following moredetailed contributions.
6
1. INTRODUCTION 1.3. Dissertation Outline
1.2.4.1 Process Models’ Quality Verification (CTA)
A problem found (see RPA) in process modeling with BPMN is the need to check well-formedness of models according to the rules of the standard, in order to achieve a goodquality artifact.
There are three types of rules that can be considered when checking models’ correct-ness: (1) well-formedness rules defined at the BPMN standard document in plain text(section 5.2.1); (2) best-practices rules promoted by BPMN practitioners in the literature(section 5.2.2); and (3) rules regarding formal properties (e.g. deadlock, liveness) enforcedby available model checkers3.
In this dissertation, we will add to BPMN metamodel the two first types of well-formedness rules for enforcing the process diagrams’ quality.
1.2.4.2 Process Models’ Quality Measurement (CTB)
Another problem (see RPB) found in process modeling with BPMN is the lack of mea-sures that can help process modelers to be knowledgeable of BPMN models’ internalattributes (e.g. size, complexity). These measures could give process modelers hintsand guidelines for achieving business process model with superior external quality char-acteristics (e.g. correctness, understandability, maintainability). Generally, appropriateinternal properties of a process model are a pre-requisite for achieving required externalqualities [ISO05b].
In this dissertation, we will derive and add to BPMN metamodel measures for as-sessing BPMN models’ internal attributes and thus contribute for enforcing the businessprocess diagrams’ external qualities.
1.3 Dissertation Outline
The control-flow, as well as the main data exchanged among of the activities of this disser-tation (chapters of this document), was organized according the process model illustratedin Figure 1.1.
1. Inception Phase:• chapter 1 – In this chapter, we presented the motivation for the subject of qual-
ity on process modeling. Since we chose BPMN for expressing the qualityin process models, we define our research problems based on the fact thatBPMN has limitations for the design of good quality models. Backed up bythe elicited research problems we formulated the research questions and thethesis statement of this dissertation. The expected contributions from the re-search work were also presented.• chapter 2 – We addressed the subject of processes and process modeling, since
their early inception to the current days. We highlighted several engineering
3This kind of rules is already covered by several proposals and tools (see chapter 4).
7
1. INTRODUCTION 1.4. Conclusion
approaches that use the process paradigm, and introduced also, in more detail,the subject of quality upon process modeling. We assessed relevant processmodeling languages, and justified the choice of BPMN for studying the qualityregarding process models.• chapter 3 – We analyzed the BPMN, as well as the relationship between the
BPMN metamodel and the graphical constructs of the language. Furthermore,by surveying a set of BPMN tools, we gathered evidences of the limitationsof the BPMN, concerning the formalization of well-formedness rules, whichhinders the production of good quality models.• chapter 4 – We surveyed the state-of-the-art of previous attempts to validate
BPMN models, highlighting the merits and limitations of those approaches.2. Research Phase:
• chapter 5 – Our approach to overcome BPMN limitations is introduced, by de-riving and formalizing the well-formedness rules for enrichment of the BPMNmetamodel.• chapter 6 – A proposal of a set of measures is formalized for measurement of
BPMN models’ internal and external quality characteristics.3. Implementation Phase:
• chapter 7 – An MDE approach was introduced for instrumentation of processmodels’ transformations and the data collection for the experimental studiesto be presented in the next two chapters.
4. Validation Phase:• chapter 8 – Using a research approach based in the scientific method, two ex-
perimental studies are presented to give empirical evidence about the effec-tiveness of the verification approach proposed in chapter 5.• chapter 9 – Similarly to the previous chapter it is also presented here a experi-
mental study. The aim is to support, through empirical evidence, the measuresproposed in chapter 6 regarding process models characteristics, as well as therelationship between models’ faults and those characteristics.
5. Conclusion Phase:• chapter 10 – Some conclusions resulting from the research work previously
done are discussed. The future work envisaged is also summarized.
1.4 Conclusion
This chapter was intended to give the context of the dissertation, by paving the wayfor the process modeling overview (chapter 2), the recent developments regarding pro-cess modeling and the BPMN (chapters 3 and 4), the presentation of main contributions(chapters 5 and 6), as well as the results’ validation (chapters 8 and 9).
The chapter begins by providing the motivation for the research work on quality ofprocess modeling (section 1.1). The chosen process modeling language (BPMN) revealed
8
1. INTRODUCTION 1.4. Conclusion
some weaknesses that affect the quality of produced process models. This evidence set-tled down the framework of the research work, through the definition of the researchproblems, the research questions, the thesis statement and the main contributions ex-pected from the dissertation (section 1.2). Ultimately, a plan for the dissertation wasoutlined to accomplish the proposed research objectives (section 1.3).
9
1. INTRODUCTION 1.4. Conclusion
Figu
re1.
1:Pr
oces
sm
odel
ofth
edi
sser
tati
on
10
2Process Modeling
"Never invest in a business you can’t understand."
Context: Nowadays, the process paradigm embodied by business processes is part ofall organizations. Process modeling is concerned with the representation of processesthat take place within and between organizations, by means of process models.Objective: Determine the relevant factors, a chosen process modeling language mustpossess, for addressing the quality of process models.Method: A set of process modeling languages was assessed for choosing the most ap-propriate to tackle the issues of quality of process models.Results: BPMN was chosen for addressing the quality of process models.Limitations: The survey on process modeling languages considers only the set of mostrelevant process modeling language.Conclusion: BPMN is nowadays the most well equipped process modeling language.
11
2. PROCESS MODELING 2.1. The Process Paradigm
2.1 The Process Paradigm
The Industrial Revolution, in the late 18th and early 19th centuries, brought a fundamentalnumber of innovations that led the boost of unprecedented economic and social changes.This came mainly from the transition to new manufacturing processes, going from handproduction to machine based methods, underpinned by new organization of human la-bor in factories through assembly lines, which greatly improved efficiency [Hob81]. Sub-sequent studies from Frederick Taylor [Tay11], among others, focused on work simplifi-cation, for time-motion, systematic experimentation to identify the best way of perform-ing a task, and control systems that measured and rewarded output. Were also relevantthe efforts of Henry Ford building standardized parts and assembly lines, which drewattention on the systemization of industrial and organizational tasks, to optimize the effi-ciency of departments and the overall organization [WM08]. These historical antecedentspaved the way for the ulterior economical and organizational transformations witnessedin the 20th century, supporting products delivery, wrapped up by services, as well as theprimacy of dematerialized transactions in a society driven by information [LL05].
In the beginning of the second half of the 20th century, computer applications stealthilystarted to be used in organizations to support the flow of activities. However, due tomethodological and technological constraints, programmers were faced with the neces-sity of coding, in each application, all the basic functionalities (e.g. access to persistentstorage and memory management). As a consequence, these functionalities redevelopedfrom scratch for each new different application, became embedded in each application,tighten up the application to the specific department for which it was developed [Wes07].This led to information systems made up by isolated applications supporting organiza-tional functions – the so called silo-based applications [Har07]. Since those software sys-tems were owned by specific organizational units, the regular business1 process flowsand information sharing was hampered, both inside the organization and with externalpartners. This situation had negative effects on the ability of organizations to react tochanging requirements, induced by a dynamic market environment, as well as changesin technology, and regulatory legislation, which had to be reflected in software systems.
The evolution in software development, due to the incorporation of principles (e.g.separation of concerns [Dij82], information hiding [Par72]) and methods (e.g. require-ments engineering, object-oriented analysis and design), alongside technological inno-vations, contributed to relieve information flow from the restraints brought by earliercomputer systems. Database management systems, for instance, in the decades of 70 and80 of the 20th century, released applications from data constraints by guaranteeing theprinciple of data independence [Cod70] and the primacy of data modeling [Che76].
Enterprise Resource Planning (ERP) systems in the 90s brought a new way of deliv-ery to applications systems. The emphasis shifted from programming to assembling and
1In this dissertation we use the term business not in sensu stricto of a commercial activity, but in a sensu latoof an activity that someone is engaged in (see the Oxford Dictionary at http://oxforddictionaries.com).
12
2. PROCESS MODELING 2.1. The Process Paradigm
configuring systems’ functionalities, by consolidating disparate applications with an in-tegrated corporate database. However, additional business requirements demanded newtypes of software systems such as, Supply Chain Management (SCM) and Customer Re-lationship Management (CRM). Ultimately, these systems led to situations previouslyseen such as data redundancy and lack of data consistency of siloed applications. Due tothe increased complexity of those systems, Enterprise Application Integration (EAI) hadto be used to integrate data in heterogeneous information technologies.
EAI applications also revealed some drawbacks since, to a large extent, they rely onprogramming and configuration of adapters and message brokers. Typically, what infact was embedded in an EAI system was the sequence of activities to be executed, con-strained by a set of rules that implemented a business process2 aiming a specific businessgoal. Therefore, in EAI solutions, the process model was hard-coded into the informationsystem [Wes07].
Workflow Management Systems (WfMS) proposal to solve the process integrationproblem was to promote business processes to first-class entities. A Business Process(BP) can be defined as a timely and space orchestrated set of activities, carried out byan organization, either internally or in conjunction with other organizations, to attain anoutput of value for stakeholders [Dav93, HC93].
WfMS effectiveness intends to be based on configuration of information system throughbusiness process models. An higher flexibility of information systems is aimed sincechanges dictated by new requirements are introduced via process models [vvdAS11].Using WfMS analysts can define and execute business processes and the rules governingprocess decisions [Har07]. Another non negligible benefit of WfMS usage was the accessto documentation of business process, for compliance with regulatory purposes (e.g. au-dits of corporate IT processes to follow the Sarbanes-Oxley Act (SOX) [Ko09], and theThird Basel Accord (BASEL III) [Sup10] that financial institutions have to comply with[Har07]).
Most WfMS vendors follow a Workflow Reference Model created by the WorkflowManagement Coalition (WfMC) for defining WfMS’s components and functionalities aswell as to identify its most important interfaces (Figure 2.1), to facilitate exchange of in-formation in a standardized way, thus enabling interoperability between different prod-ucts [CT12]. The information systems that interface with WfMS, by executing businessprocesses involving either people, applications, or other information sources, are referredas Process-Aware Information System (PAIS) [DVDATH05].
In Figure 2.1, the component with dash border (Process Definition Tools) is the mostrelevant for the current dissertation, since it is the one that supports activities of business
2In this dissertation when the concept of business process is mentioned it does not mean that we arerestricting the scope of this work to commercial activities in organizations. The concept of business processmust be considered in sensu lato, i.e., regarding any kind of process composed by manual or automatedactivities that can be done in parallel, and thus must be synchronized, consuming some sort of resources(products or services) and generating an output of value (product or service) for a stakeholder. We leaveoutside this definition, and so outside the scope of this work, all kind of industrial processes with real-timerequirements.
13
2. PROCESS MODELING 2.1. The Process Paradigm
process modeling, which is our main focus in this work.
Figure 2.1: The Workflow Reference Model and main associated standards (Sources:WfMC/OMG/W3C)
Currently, the most used software tools for supporting business processes are Busi-ness Process Management Systems (BPMS) [HCKP09]. They became a step forwardregarding WfMS, by incorporating a diagnosis phase in the business process life cycle.In the diagnosis phase, process analysts are able to identify and improve on bottlenecksand potential fraudulent breaches of business processes using the appropriate analysisand monitoring tools [Ko09]. Furthermore, BPMS take advantage over WfMS on thecontemporary distributed environments of Web services and Service-Oriented Architec-ture (SOA) [Erl07].
The adoption by BPMS of SOA’s IT architectural paradigm, the latest computingparadigm incarnation of Service-Oriented Computing (SOC), allow the use of technol-ogy with greater agility, with computational processes linked (e.g. by web services) inorder to enable the coordination of distributed systems that support processes. So, busi-ness functionalities, can be encapsulated within services and made available in reposito-ries through interfaces and message protocols. Therefore, conversely to the WfMSs thatwere based on the WfMC’s idea of a centralized enactment engine confined to a singleorganization, the BPMS business processes can be implemented via services’ composi-tion upon the SOA technological infrastructure. With SOA, it becomes easier to composeand maintain information systems to the point that it is possible to shift from a care-fully planned design of information systems to an on-going and permanent redesign andorganic growth [vvdAS11].
With Software as a Service (SaaS) adoption, organizations do not need to host and runtheir own applications. They can subcontract applications’ usage to external suppliers.Even the infrastructure, can be virtualized and shared among other organizations when
14
2. PROCESS MODELING 2.2. The Business Process Perspective
supported by cloud computing. Since an unified and integrated perspective of organiza-tions’ information systems can be provided by a business process perspective, no wonderthat BPMS became more and more, an important tool to cope with the later technologicaltrends, in search for an increasingly support and automation of processes.
2.2 The Business Process Perspective
Processes are a set of coordinated activities and procedures fueled by resources, aimingto fulfill specific goals of particular stakeholders. In this dissertation, we are concernedwith the modeling of specific type of processes: business processes3 (e.g., purchase or-ders, shipping management, claim register, etc.), which are a blended set of automatedand manual activities that take place in organizations. Business processes constitute themeans through which organizations seek to achieve their main objective of satisfying theclients’ demands by delivering goods or services [MaP01].
A process can be formally documented, or merely informally defined. Regardless itsnature, processes exhibit features that make them suitable for the usage of modeling tech-niques. Business Process Modeling (BPM)4 is the set of activities conducted for visuallydepicting qualitatively grounded models of organization’s processes, so that processescan be analyzed, monitored and improved regarding their expected value. Through pro-cess modeling the As-Is processes are analyzed to become improved To-Be processes inthe future.
Business process modeling is a cross-disciplinary research area that adopted a widevariety of paradigms and methodologies from different areas such as organization man-agement theory, computer science, mathematics, linguistics, semiotics, and philosophy[Ko09]. The aim of business process modeling5 is building Business Process Diagrams(BPD) the abstract representations of processes, using either technical drawings, depict-ing network of graphical elements [Sin06], or structured textual information serializedin eXtensible Markup Language (XML) derived formats (e.g. XML Process DefinitionLanguage (XPDL), XML Metadata Interchange (XMI)).
Since process diagrams are intended to support the activities of stakeholders withdifferent technical backgrounds (e.g. process analysts, process designers, process imple-menters), they should be suitable, among other purposes, to facilitate the communicationamong those kinds of stakeholders [Hol09].
Business process modeling is part of a multidisciplinary research domain, the Busi-ness Process Management6, which comprises a set of activities that besides modeling,
3Other type of processes are referred in section 2.3, more related with industrial control system andsystems with real-time requirements, therefore outside the scope of this dissertation. To not overload thetext we will replace whenever possible the term business process simply by process.
4The term was coined in the 1960s in the field of systems engineering [CT12].5To not overload the text we will replace whenever possible the term business process modeling simply by
process modeling.6The acronym BPM, used previously for Business Process Modeling, is also shared by the Business Pro-
cess Management community which sometimes originates that one knowledge area is confused with the
15
2. PROCESS MODELING 2.2. The Business Process Perspective
includes the management, execution and analysis of processes [VdAtHW03]. BusinessProcess Management is a set of methods, techniques and software, related to the sup-port of processes’ design, enactment, control and analysis, involving several sources ofinformation such as humans, organizations, applications, and documents.
Business Process Management is a practical, iterative, and incremental approach7 offine-tuning business processes, leveraging IT [Ko09]. Business Process Management isnot a technology and simply provides governance of a business’s process environmentto improve agility and operational performance using a systematic approach [CT12].
Business Process Modeling (BPM) is the cornerstone in the life cycle of business pro-cess management and is involved in all its phases [WHJC06]. The Business Process Man-agement life cycle phases, as well as its contribution, are summarized below [VdAtHW03,WHJC06]:
1. Design – the business process diagrams are captured from the actual processes (As-Is business process model) and stored using graphical tools. Business process mod-eling is generally initiated by process analysts and managers, pursuing efficiencyand quality of business processes.
2. Configuration – the BPMS and the underlying system infrastructure is customizedto the organization’s specificity. The business process diagrams are adjusted to fitthe actual environment of the operations.
3. Enactment – business process diagrams are deployed in a BPMS, and become exe-cutable in a real environment.
4. Diagnosis – analysis and monitoring tools assess business processes execution basedon the business process model. The results concerning advantages and shortagesof current business processes are displayed, preferably using a representation witha high level of abstraction. Further improvements are assessed and another turn ofthe cycle might be triggered.
The design of processes through process modeling is the activity carried out in theearly stages of a process elicitation. Tackling process modeling helps to identify prob-lems in the early phases of process development [RCG+09] and assists the design ofvalid process models. Moreover, a suitable analysis and verification of processes at themodeling stage would make easier the process maintenance tasks [ARGP06] by reducingits implicit costs. That is why, as previously noted regarding requirements engineeringand software development [SD97], it pays off the effort made upon the conceptual levelof producing valid processes. Therefore, it is relevant to impose quality characteristics(e.g. correctness, understandability, maintainability) to process diagrams.
other.7The previous approach, enhanced and overcame by Business Process Management, was the Business
Process Re-engineering (BPR) [HC93, Dav93], which took a radical reshaping of existing business processes,through the obliteration of forms of work that did not add value to the organization.
16
2. PROCESS MODELING 2.3. Other Process Engineering Approaches
Figure 2.2: The Business Process Management life cycle (Adapted from: [Wes07])
2.3 Other Process Engineering Approaches
Modern societies progress is powered by processes [HCV10]. Besides business process,previously mentioned (section 2.2), exist other engineering approaches, for which pro-cesses play a pivotal role [VdAVH96]. We highlight some of the approaches whosetechniques (e.g. graphical notation, simulation, and performance evaluation), as wellas research efforts and practitioners experience, have been influential and benefited thedevelopment and maturity of the relatively new multidisciplinary research area of busi-ness process management in general and business process modeling in particular. Wehave been witnessing to a confluence and integration among these different process per-spectives [Har04], for which, the sharing of a common graphical representation wouldcertainly give a greater contribution.
2.3.1 Process Engineering
Process Engineering includes, among others activities, process design, process control,and process operations. Process design consists in the design of new products or in themodification or expansion of existing ones. Process models, made at the conceptual level,serve to define and ensure that the design components fit together at the end of fabrica-tion and construction plans. Some of the modeling techniques used in Process Engineer-ing are: Block Flow Diagrams (BFD), Process Flow Diagrams (PFDs), and Piping andInstrumentation Diagrams (PIDs) [GW00].
Process Engineering applies systematic computer-based methods on the design, oper-ation, control, and optimization of biological processes, chemical, and physical, [OM98].This includes a wide range of industries, namely food, pharmaceutical, chemical, mineralprocessing, petrochemical, advanced material, and biotechnological [GH13].
17
2. PROCESS MODELING 2.3. Other Process Engineering Approaches
2.3.2 Systems Engineering
Systems Engineering is an interdisciplinary field of engineering, focused on the designand management of complex engineering projects and systems over their life cycle [KSSB11].These systems may include hardware, software, data, personnel, procedures, and facili-ties from a broad range of industry domains (e.g. aerospace, automotive, health care).System Engineering aims to analyze, designing, and organizing elements into a sys-tem that can be a product, a service, or a technology for the transformation of infor-mation [PI92]. The practice of Systems Engineering underwent by a key transition froma document-based approach to a model-based approach. In the model-based approach,the focus shifts from documentation production and control to an integrated model ofthe system that could help to manage complexity [FMS11].
Among the system modeling techniques used by Systems Engineering that can beconsidered predecessors of the current business process modeling techniques, we high-light [KSSB11]: Functional Flow Block Diagram (FFBD), Data Flow Diagram (DFD), andIDEF0 Diagram [WP93]. Recently, as a response to the requirements issued by the OMGto extend UML to support systems modeling, the Systems Modeling Language (SysML)[OMG12b] came to light. SysML is a graphical general-purpose modeling language thatsupports the specification, design, analysis, and verification of systems. The SysML se-mantic foundation allows for the representation of requirements, behavior, structure, andproperties of the system and its components [FMS11].
2.3.3 Industrial Engineering
Industrial Engineering is a branch of engineering concerned with the improvement, de-velopment, implementation and monitoring of integrated systems of people, informa-tion, equipment, and material, in order to attain high quality products manufactured atlower costs, with shorter cycle times. The Industrial Engineering field usually overlapswith many other knowledge areas (e.g. management science, systems engineering, op-erations management, operations research, ergonomics or human factors engineering,manufacturing engineering, and safety engineering) regarding specification, predictionand evaluation of systems implementation results. The decision-making process is sup-ported by software process modeling, expert systems, business process re-engineering,simulation software, costing manufacturing models, and several other methodologiesand tools [HMVHR06].
2.3.4 Quality Engineering
Quality Engineering deals with the analysis of manufacturing systems throughout prod-uct or services life cycle, to improve the quality of the production process, as well as itsoutput, in order to satisfy customer expectations at individual level [HBS00]. Quality en-gineering tools and techniques include among others, SPC (Statistical Process Control),DoE (Design of Experiment), Taguchi methods and QFD (Quality Function Deployment).
18
2. PROCESS MODELING 2.3. Other Process Engineering Approaches
Quality Engineering is tightly related with quality frameworks, such as Total QualityManagement (TQM), Six Sigma and Lean:
• With TQM the focus is the satisfaction of agreed internal and external customers’requirements [Jon94]. TQM implementation rely on the deployment of a Qual-ity Management Systems (QMS), as defined in the ISO 9000 family of standards[ISO05a], to assist organizations to handle a permanent monitoring of the cus-tomer’s requirements as well as the continuous improvement of the organization’sprocesses and services [Lon04]. ISO 9000:2000 promotes the use of a process ap-proach when developing and structuring the various QMS activities of an orga-nization through ISO 9004:2000 [ISO00] implementation, namely by highlightingthe importance of: (i) specifying quantitatively requirements and their accomplish-ment; (ii) considering processes in terms of measurable added value; (iii) gettingresults of process performance and effectiveness; and (iv) continual improvementof processes based on objective measurements.• Originated also from manufacturing processes, Six Sigma seeks the improvement of
the quality of processes’ outputs by identifying and removing the causes of defects(errors) and minimizing variability in manufacturing and processes, grounded onstatistical modeling [Per06, Bru04].• Lean is a systematic methodology to reduce the complexity and streamline pro-
cesses by identifying and eliminating sources of waste that typically causes ineffi-ciencies in the processes’ flow [Wed06].
2.3.5 Performance Engineering
Performance Engineering is a discipline related with computer systems that relies onstatistics, queuing theory and probability theory. Performance engineering uses toolsand processes to ensure that a system, throughout its life, meets the customer’s expec-tations for performance, using namely analytical and simulation modeling, as well asperformance testing [MADD04].
2.3.6 Software Engineering
In Software Engineering it is generally accepted that a quality development process is in-evitably more likely to produce quality software consistently than an ad hoc developmentprocess [Cro00]. The quality is correlated to the maturity of an organization’s softwareprocess on a predefined quality scale. Currently, there are three important initiatives whodeal with the definition and quality measurement of software processes in particular, andprocesses in general [KK97a]:
• the Capability Maturity Model Integrated (CMMI) [SEI10c, SEI10b, SEI10a], intro-duced by Software Engineering Institute (SEI) as a generalization of Capability Ma-turity Model for software (CMM), can be applied to the wide variety of processesin different kind of organizations. The CMMI grades organization’s processes in
19
2. PROCESS MODELING 2.4. Quality in Process Modeling
an increasing scale of five maturity levels: 1 – No Organized Processes (Initial); 2 –Some Organized Processes (Managed); 3 – Most Processes Organized and Standard-ized (Defined); 4 – Processes are Measured and Controlled (Quantitatively Managed);and 5 – Processes are Continuously Improved (Optimizing).• the ISO/IEC 15504 known as Software Process Improvement and Capability dEter-
mination (SPICE) [ISO04a] derived from process life cycle standard ISO/IEC 12207and maturity models (e.g. CMM), and developed for an overall determination ofthe organization’s capabilities for delivering products (software, systems, and ITservices).• the International Organization for Standardization (ISO) produces, among others,
the series of standards ISO 9001 among which is the 9000-3 which specifies quality-system requirements for use when a contract between two parties requires thedemonstration of a supplier’s capability to design and supply a software product.
In a software development process, there is also the Software Process EngineeringMetamodel (SPEM) [OMG08b], a UML profile used to describe a concrete software de-velopment process or a family of related software development processes involving orrequiring the use of the UML [DRC+06, CdO13].
2.3.7 Other Initiatives
There are also some initiatives (e.g. Proposed Interchange Formats (PIF),Process Speci-fication Language (PSL)) aiming to join together commonalities (e.g. inputs, activities,resources, deliverables) from different modeling approaches, keeping apart the differ-ences among them (e.g. real-time aspects, models’ descriptive and executable aspects)[BB01].
Future research in business process modeling will certainly benefit from even morecross fertilization between the above mentioned scientifically more mature engineeringdomains, where the process approach has been used for decades both in industry andacademic applications.
2.4 Quality in Process Modeling
Process modeling is a design discipline that produces conceptual models (process dia-grams) from processes’ elicitation. Following a taxonomy of Mylopoulos et al. regard-ing quality attributes in software systems [MCN92], we also considerer two approachesto deal with the formal treatment of quality in process modeling: process-oriented andproduct-oriented approaches.
2.4.1 Process-Oriented Approach
The process-oriented approach to quality in process modeling focus its attention in thedefinition and assessment of the set of quality attributes that outputs of processes must
20
2. PROCESS MODELING 2.4. Quality in Process Modeling
attain. Therefore, according to this approach, alongside information regarding the char-acteristics of outputs, the processes specification should also contain information aboutthe relevant domain dependent quality characteristics (e.g. response time, availability, ac-curacy) of the provided products and services. The process-oriented approach is in linewith the critical-realist perspective [Moo05] according which, the design artifacts shouldbe used to actively construct and assess the world, rather than simply describe it. Therefore,processes’ models, as an outcome of process modeling, should also provide informationregarding the value attained by the stakeholders, i.e., the beneficiaries of the processes.
Some research threads already developed in accordance with the process-orientedapproach are described below:
• The approach of Kueng and Kawalek was innovative, addressing the business pro-cesses, modeling and assessing, at a conceptual level, namely highlighting howprocess models can be evaluated against nonfunctional goals. They concluded thatthe assessment of processes, if carried out at model level, would be always partial[KK97b].• The assessment of quality properties of processes using structural measures8 simi-
lar to those used in software [Pha98].• Korherr and List added a context perspective to an earlier work [CKO92], which
had considered functional, behavior, data and organizational perspectives. The ex-tension of some business process modeling languages metamodels with businessprocess goals and performance measures, allowed these concepts to become explic-itly visible in the corresponding models. Furthermore, they propose a mappingof the performance measures in process modeling languages onto the process ex-ecution languages intended to enable the monitoring of process at the executionenvironment [LK06, KL07, Kor08].• Go4flex approach [BPJ+10] uses process goals to describe what is to be achieved, in-
stead of addressing exclusively the activities that should be done, using traditionalactivity oriented workflow languages. They proposed a framework including thefive perspectives from Korherr and List, as well as concepts developed in the areaof agents and multi-agent systems.• Popova and Sharpanskykh’s approach, coming from the areas of Artificial Intelli-
gence (AI) and Business Engineering (BE), proposed an approach of a model of anorganization that included its goals, as well as relevant Performance Indicators (PIs)(aka measures) for measurement and analysis of the organizational performance.The contribution of the approach is a framework, with formal foundations, whichallows the formalization of the PI concept, the relationships among PIs, as well asthe analysis and verification of temporal aspects of the represented relations, withthe language Temporal Trace Language (TTL) [PS08, PS09, PS10a, PS10b, PS11]
8We use the term measure instead of the term metric for reasons that are detailed in section 6.2, namelybecause its semantic overload, which cause that more recent standards (e.g. ISO/IEC 25000 and ISO/IEC15939 series) disregarded the use of the term metric.
21
2. PROCESS MODELING 2.4. Quality in Process Modeling
• In [Joh08] the UML Schedulability, Performance, and Time (SPT) specification [OMG05]was composed with the UML Activity Diagram (AD) specification [OMG07a] to al-low a language for simulation of processes execution within a process diagram. In[BD12] it was introduced a model-driven method for performance prediction of au-tomated processes, implemented as an orchestration of web services, based on anBPMN extension for the specification of performance properties.• In [PZ08] it was proposed to be added to BPMN symbols, new artifacts to model
process models constraints and operational conditions. The method intended toassist the early discovery of non-functional requirements during systems develop-ment. A tool support for the method was made available in [HKP11b]. The sameauthors also presented a survey regarding modeling quality information withinprocess diagrams in [HKP11a].
2.4.2 Product-Oriented Approach
The product-oriented approach attempts to define methods and tools to ensure that processmodeling outcomes, i.e the process diagrams, can be evaluated to the degree to whichthey meet certain external quality attributes (e.g. correctness, understandability, main-tainability).
To evaluate the quality of process diagrams, as with the software artifacts, we mustbe aware that quality is a multi-dimensional concept, with several characteristics. TheISO/IEC 25000 series (Software product Quality Requirements and Evaluation – SQuaRE)[ISO05b], the successor of ISO/IEC 9126 standard, is a well known example where thefactors contributing to quality are grouped to constitute the basis of a quality model. Aquality model often decomposes the view of quality into different levels of quality char-acteristics. The inter-relationships between quality characteristics and their measures arealso documented. Eventually, the internal attributes are linked to external quality at-tributes. In the case of process modeling, this results in a quality model with directlymeasurable attributes of process models diagrams (e.g. size, complexity) linked to theperceived quality by stakeholders (e.g. correctness, understandability, maintainability).So, since a quality model is a framework for quantifying and linking different qualitycharacteristics, one could expect that if we are able to control the internal quality char-acteristics of process models, this will ensure more control over the final quality of themodeling process.
For the sake of focus, in this dissertation it will be given more attention to the product-oriented aspects of process modeling. However, given the relevance of process-orientedperspective, it will be also briefly addressed as future work in chapter 10 (section 10.3).
22
2. PROCESS MODELING 2.5. Process Modeling Languages
2.5 Process Modeling Languages
For assessing the quality characteristics of the outcomes of process modeling, we needto do it through the elements of the process modeling language. Several notations havebeen developed to represent process models, since the introduction of flowcharts in the1920s [IMR09]. Those notations are known generally as Business Process Modeling Lan-guage (BPML). Depending on the particular BPML, different syntactical and semanticconstructs are used to represent concepts regarding processes, namely actions, flows ofcontrol, data, or resources.
BPMLs offers a wide range of characteristics and functionalities, more or less busi-ness or IT oriented. Hence, one can find some BPMLs presenting a rich set of graphicalconstructs, inspired in the language terminology used in the business domain, while oth-ers, emphasize aspects such as the formal semantics, analysis methods and executionenvironments.
Given the diversity and demanding requirements, an assessment is needed to choosethe suitable BPML for each specific job, such as the measurement of quality of processmodels. Several surveys studying, comparing and positioning BPMLs came to lightalong the last decade (e.g. [Gia01, VdAtHW03, WHJC06, LK06, LS07, RRIG09, MTJ+10]).
In line with [MTJ+10], BPMLs can be coarsely classified in the following two cate-gories:
• semi-formal: languages that share concerns for understandability, and are amenableof various informal or heuristic analysis. As examples of semi-formal BPML weconsider BPMN [BPM11], UML Activity Diagram (AD) [OMG07a], Event-drivenProcess Chain (EPC) [SN00]; and• formal/executable: languages that are formal (both syntax and semantics are pre-
cisely defined9) and/or executable (describing the flow of activities through scriptsthat can be executed by workflow engines). Examples of these kind of BPMLs areBusiness Process Execution Language (BPEL) [OAS07], Petri Nets (P/N) [Pet62,Mur89], Yet Another Workflow Language (YAWL) [tHvdAAR09], and Subject-oriented Business Process Management (S-BPM) [Fle10].
In this work we do not address BPMLs specially tailored for dealing with specificproblems [MTJ+10] namely: (1) process integration in electronic commerce solutions, bringtogether different business partners through abstract and technology independent pro-gramming interfaces and data exchange formats (e.g. RosettaNet, ebXML, BPEL4WS);and (2) object-oriented modeling languages, more adapted for representing the software(solution domain) rather than the business (problem domain) (e.g. EDOC). A brief cata-log of other BPMLs not covered in this work is presented in Appendix A (section A.1).
To address the research topic of quality in process modeling, we needed to choose,
9When the semantics is formally defined, sentences in the language have a unique interpretation, i.e. thesemantic of process instances resulting from process models specified in the language is well defined andnot ambiguous [Wes07].
23
2. PROCESS MODELING 2.5. Process Modeling Languages
according to certain criteria, a language suitable for the specification and assessment ofthe quality attributes in process models. For this purpose, a pragmatic taxonomy is usedin this research work to enable a grounded justification of the choice made.
2.5.1 A Taxonomy for BPMLs
A taxonomy is a particular classification, usually organized by generalization-specializationrelationships (aka inheritance relationships). At the top of the taxonomy structure (rootnode) there is a single classification that aggregates common characteristics to all an-alyzed languages – in this case the root node is labeled as Process Modeling Language.Nodes below the root node are more specific classifications that are applied to subsets oflanguages. The assignment of each language to a criterion node was based on an ordinalscale. Next it will be detailed each criterion of the taxonomy, as well as the underlyingordinal scale criteria:
1. Extent of Concepts: is related to the breadth of the concepts in the process modeling lan-guage, which is related to the extent of the use of the modeling language (e.g. aimingto be used only with one purpose or several purposes, such as processes’ documen-tation and communication instrument, simulation and execution). Several constructsclose to the domain concepts allow the language to have a broader scope, thus making itamenable to be used by process analysts and process implementers. On the other hand,a language with a small number of constructs, such as a general-purpose language thatincludes constructs designed to be used in several contexts and not exclusively within theprocess modeling, will have limited scope and, thus, will be less effective in conveyingthe specificities of modeling situations.
(i) Narrow: the language has a small number of primitive constructs (up to 20), whichhowever can compose complex concepts. The language extent is limited to a specifickind of task (e.g. modeling, implementation or verification). A possible drawbackof the language could be the complexity of the attained process models, as well asthe language’s learning curve to users;
(ii) Intermediate: a medium number of constructs (20 to 40) are made available to expressprocess models. Due to the fact the language uses concepts that are closer to theprocess domain, the semantic gap among different communities of users is lower.However, the language extent is also limited to a short set of tasks (e.g. modeling,simulation, verification) within process modeling;
(iii) Broad: a larger number of constructs (more than 40) are available for process model-ing. Simpler models can be achieved given the semantic richness of the constructs.The extent of the language is broader, since it could include tasks such as modeling,implementation and verification.
2. Level of Adoption: is related with the dissemination of the language as process modelinglanguage.
24
2. PROCESS MODELING 2.5. Process Modeling Languages
(i) Scarce: used only in restrict communities, given the level of expertise required tomaster the language (e.g. researchers), or the fact of being attached to specific tools;
(ii) Moderate: the diffusion spread out several communities with different interests re-garding process modeling;
(iii) Wide: massive adoption to the point that became a de facto standard.
3. Target Users: is related to the main beneficiaries of the language. The main users wouldbe among the members of academia and industry. Ideally a language should be able toserve activities of design, documentation, validation and verification, regarding processmodeling, on both communities.
(i) Specific Community: the language has a great level of specialization and it only in-tends to serve a specific community (academics or practitioners);
(ii) Some Users of Several Communities: albeit the language addresses both communities,only specific type of users in each community are targeted, given the specificity oftheir activities;
(iii) All Users from Several Communities: besides addressing both communities, users withseveral technical backgrounds, in each community, benefit from the use of the lan-guage. For instance in the case of practitioners: process analysts (for process elici-tation and design) and process implementers (for process verification and deploy-ment in BPMS engines).
4. Depth in Verification: it concerns to the extent and deepness of verification process avail-able directly from the language itself or via transformation to another language or tools(e.g. properties checking tools).
(i) None: there is no concern about verification
(ii) Semi–formal: the emphasis goes to the syntactical verification of process models;
(iii) Formal: ensures a syntactical and semantic checking, as well as properties verifica-tion of process models.
5. Kind of Notation: is about the symbolic representation used by the language to representprocess modeling constructs
(i) Diagrammatic: the approach emphasizes the graphical notation;
(ii) Mathematical: the approach is grounded by logical or algebraic foundations, target-ing processes’ simulation, execution or formal verification;
(iii) Diagrammatic & Mathematical: a composition of graphical and mathematical nota-tions provides both usability and soundness to process modeling.
6. Tools Availability: it is related with the users’ accessibility to a diversified set of toolssupporting the language for process modeling, as well as transformation to another for-mat to benefit from models’ interchange among tools. A language being supported by a
25
2. PROCESS MODELING 2.5. Process Modeling Languages
vast range of tools, enables its widespread adoption and practice usage. Furthermore, theavailability of a large number of tools, increases the competitiveness among tools mak-ers for enhancing the tools’ functionalities, anchored in research and innovation, whichultimately benefits users.
(i) Few: only a small number of tools and methods for that language, are available (upto 10);
(ii) Some: a larger number of tools and methods (10 to 30 tools) with similar character-istics, ensures some choice options to users;
(iii) Many: a competitive market (more than 30 tools) shortens the life cycle of tools’ ver-sions. New available releases bring new functionalities and enhancements regard-ing the process modeling. Are available also, alternatives for models’ interchange.
2.5.2 Characterization of BPMLs
In the following sections are analyzed and ranked the set of above mentioned semi-formal and formal/executable languages, according to the specified taxonomy. We areaware that these languages were developed with different audiences and objectives inmind, but all of them have been used to describe different facets of processes. The in-tent of the analysis is not to explore the construction rules for process models, using eachBPML. Rather we intend to analyze the BPMLs, comparatively and critically, along eachdimension of the taxonomy. Below we list all the analyzed BPMLs, as well as the sectionwhere the analysis is detailed.
• Business Process Model and Notation (BPMN) (section 2.5.2.1)• Activity Diagram (AD) (section 2.5.2.2)• Event-Driven Process Chain (EPC) (section 2.5.2.3)• Petri Nets (P/N) (section 2.5.2.4)• Yet Another Workflow Language (YAWL) (section 2.5.2.5)• Subject-oriented Business Process Management (S-BPM) (section 2.5.2.6)• Business Process Execution Language (section 2.5.2.7)
2.5.2.1 Business Process Model and Notation (BPMN)
1. Extent of Concepts. BPMN was developed to enable business users to build readilyunderstandable representations of business processes. BPMN is one of the most recentBPMLs, so it is grounded on the experience of earlier process modeling languages, whichontologically makes it one of the most complete BPMLs [RIRG05, RRIG09]. BPMN canbe used within many methodologies and for many purposes, from high-level descriptivemodeling to detailed modeling intended for process execution [Whi05]. The Object Man-agement Group (OMG) made the BPMN 2.0 version more technical and more IT oriented,so besides the graphical notation (see Figure 3.4), BPMN encompasses a level of detailregarding modeling constructs, which enables BPEL code generation. So, by supporting
26
2. PROCESS MODELING 2.5. Process Modeling Languages
the details of graphical object’s properties, BPMN offers a standardized bridge for thegap between the business process design and process implementation [Whi05]. (Ratingaccording the taxonomy: Broad.)2. Level of Adoption. BPMN has emerged as an important open standard notation fordepicting and modeling business processes [Gao06]. In a survey dated from 2011, in-cluding 559 practitioners [HW11], when asked which language they were using, the vastmajority (72%) acknowledged the use of BPMN. Relying on the mentioned survey, onecan conclude that, by far, the most relevant process modeling language nowadays forpractitioners, is the OMG’s BPMN standard. (Rating: Wide.)3. Target Users. The focus of BPMN is producing understandable documentation ofbusiness process models, to enable business analysts to develop understandable graph-ical representations of business processes that can be used by different kinds of usergroups [BPM11]. So, target beneficiaries of BPMN range from the process developer,who has more focus on the technical aspects, up to the business analyst, who is focusedinto managing and optimizing of business processes. There are also plenty of researchworks in academia about BPMN, from ontological and usability, formalization to imple-mentation [RIRG05, BO10, DDO07, REH11, Whi05]. (Rating: All.)4. Depth in Verification. BPMN language is specified by means of a metamodel [BPM11]expressed with the UML, the notation which is the de facto standard for modeling soft-ware engineering artifacts [OMG07a, OMG07b]. BPMN metamodel enables the syntacti-cal validation of business process models by tools. Furthermore, several transformationsare also available, from BPMN to other languages, complementing the models’ syntac-tical validation with properties formal verification, using notations and specific modelchecking tools such as Petri nets [DDO07], Communicating Sequential Processes (CSP)[WG08], and Colored Petri Nets (CPN) [REH11]. (Rating: Semi–formal.)5. Kind of Notation. BPMN has a semantical rich graphical notation (see Figures 3.2an 3.3). Besides the regular constructs for activities’ control flow behavior, BPMN offersa visual representation for abnormal flows, such as fault handling, escalation handling,and compensation handling. BPMN also supports data flow and interaction behavior.BPMN is considered amenable for being used in processes’ elicitation within companies,as well as representation of processes’ interactions among organizations [Tsc10]. (Rating:Diagrammatic.)6. Tools Availability. BPMN, with more than 80 diagrammatic tools10, is nowadays thebusiness process modeling language with more tool support available. BPMN tools gen-erally allow for modeling artifacts being interchanged among tools from different ven-dors. Most of the interchanged models use a serial representation of business processmodel through dialects of XML11: (1) XPDL defines a common interchange format; theBPMN and the XPDL specifications address the same modeling problem from different
10See a list in http://www.bpmn.org/ [accessed in Feb. 10, 2013] and [HCKP09]11XML is a World Wide Web Consortium (W3C) standard available in http://www.w3.org/XML/ [ac-
cessed in Feb. 10, 2013].
27
2. PROCESS MODELING 2.5. Process Modeling Languages
perspectives (graphical notation vs. text notation); (2) XMI is a model driven XML in-terchange framework for definition, manipulation and integration of XML metadata andobjects [OMG11]; XMI provides rules by which a schema can be generated for any validXMI-transmissible MOF-based metamodel [Gao06]. (Rating: Many.)
2.5.2.2 Activity Diagram (AD)
1. Extent of Concepts. The UML is a graphical notation used in object-oriented softwaredevelopment for creating visual models. The Activity Diagram [OMG07b], one of UMLbehavioral diagrams, is particularly suitable for software systems processes’ representa-tion (e.g. use cases’ description and workflow modeling, as well as methods’ specifica-tion).The UML specification provides an interrelated sets of packages specifying the features ofActivity Diagrams (see main concepts depicted in Figure 2.3). The BasicActivities pack-age provides the most fundamental features of the Activity diagram, which are thosefound in most BPMLs, such as simple activities and actions, flow from one activity oraction to another, decisions, and input and output values for actions. The IntermediateAc-tivities and CompleteActivities packages provide further refinements to this package. TheStructuredActivities package and its subpackages provide modeling support of structuredconstructs found in traditional programming languages, such as loops and conditionals,as well as exception handling.Activity Diagrams, as an enhanced version of flowcharting, addresses the functionalmodeling of systems, and was considered for long suitable for purposes of business pro-cess modeling. However, AD’s technical nature revealed ultimately that it was mostlysuited for business process automation and did not provide a comprehensive supportfor dedicated business process analysis [Lon04]. Being aware of these limitations, theOMG promoted the creation of the Business Process Definition Metamodel (BPDM)12,to handle business processes specifically. BPDM provides abstract concepts as the basisfor consistent interpretation of specialized concepts used by business process modelers[OMG08a]. (Rating: Intermediate)2. Level of Adoption. In the above mentioned inquiry from 2011 [HW11], 18% of therespondents acknowledged the use of AD for business process modeling. This seemsrepresenting a moderate to low usage of the language. The main reason seems to be theavailability of a suitable language for the purpose of business process modeling (BPMN).Therefore, AD seems to be set aside to activities more connected with software analysisand design. (Rating: Moderate)3. Target Users. The Activity Diagram has been widely accepted as an analysis anddesign technique by object-oriented communities of both academics and practitioners.However, its use remained somehow confined to the area of software development,where UML is the accepted standard [All10]. A study [BO10] comparing AD with
12http://www.omg.org/spec/BPDM/ [accessed in Feb. 10, 2013].
28
2. PROCESS MODELING 2.5. Process Modeling Languages
Figure 2.3: Activity Diagram metamodel [OMG07b]
BPMN, reveled a similar usability, as well as learning curve, by business analysts. Pos-sibly due to AD’s more process development focalization, it did not attain a generalizedand spread acceptance by business process modeling practitioners. (Rating: Several)4. Depth in Verification. Conversely to programming languages, AD lacks a universallyagreed behavioral semantics. The steps from functional specification of a system usingAD to its final implementation, faces the hindrance caused by different interpretation ofthe produced artifacts. A well-defined formal semantics of AD would allow that modelscould have a rigorous interpretation and reasoning support, as well as a being subjectto formally grounded verification. To circumvent this limitation, there are approachesseeking for formalization of AD semantics and translation to languages such as: (1) Petrinets, through the specification of the behavioral semantics for the purpose of formal ver-ification [Fah08], with the resulting Petri/Net model being amenable for verification ofcontrol-flow errors with current model checkers (e.g. LoLA13); (2) CSP gives a process’ssemantics to the AD demonstrating the consistency of the object model. CSP also facili-tates the access to other methods and tools [BD00]. (Rating: Semi–formal)5. Kind of Notation. In AD business processes are graphically represented as a directedgraph, i.e., a set of nodes and edges, with a kind of intuition inherited from Petri/Nets[EFLR98]. Some nodes denote executable action or structured activity, while others rep-resent the execution’s flow control (e.g. decisions, merges, forks, and joins). So, ADnotation represents business processes as a logic sequence of tasks governed by pointsof control. The semantics of a specific process model can be enforced by attaching Ob-ject Constraint Language (OCL) clauses to model elements, which contributes to bringmore rigor to the AD specification, thus decreasing the subjectivity inherent to the ADgraphical language. However, this is only possible at model instance level and not by
13http://www.service-technology.org/uml2owfn/ [accessed in Feb. 10, 2013].
29
2. PROCESS MODELING 2.5. Process Modeling Languages
AD construction. As with other UML diagrams, AD notation lacks a formal static andbehavioral semantics [EFLR98]. (Rating: Diagrammatic)6. Tools Availability. As a popular and mature software modeling language, UML andparticularly Activity Diagram, are widely supported by more than five dozens of toolvendors14. Since the UML metamodel is expressed by the Meta-Object Facility (MOF),XMI can be used for exchanging activity diagrams information among tools, as well asfor serialization of process models to be transformed to other languages. (Rating: Many)
2.5.2.3 Event-Driven Process Chain (EPC)
1. Extent of Concepts. Event-driven process chains [SN00] is a graphical business pro-cess language for modeling, analyzing, and redesigning business processes. EPC de-scribes processes on the level of their business logic, and intends to be easily understoodand used by business people [Kor08]. A claimed major strength of EPC notation is itssimplicity and ease of understanding [Tsc10]. EPC process models are event-driven, i.e.state-based, which means that the main focus for a process model is the representationof processes’ states and their transition, rather than the interaction and communicationamong organizations. Processes modeled with EPC depict only the internals of an orga-nization [Tsc10].The EPC is based on the concepts of stochastic networks and Petri nets. The core EPC ele-ments are formed by few constructs (function, event, OR-connector, XOR-connector, AND-connector, and control flow depicted as shadowed elements in Figure 2.4), which are un-derstood as enough for specification and documentation of process models. The basicversion of EPC was supplemented by other constructs (organizational unit, position, data,system, process link, and relation, the non-shaded elements in Figure 2.4), resulting in theextended EPC, intended to supplement process models with organizational structure anddata flow. However, EPC has less expressiveness than BPMN, and its constructs are con-siderable fewer and not so well specified as in BPMN [Tsc10].Furthermore, EPC process models are not intended for being detailed in order to be exe-cuted. EPC is a notation to model the domain aspects of business processes. The focus ofthe notation is mainly on domain concepts and processes representation rather than theformal specification or technical realization [Wes07]. (Rating: Narrow.)2. Level of Adoption. EPC existence and spread out usage as a process modeling lan-guage, was highly linked to the SAP ERP implementation. However, due to the fact ofbeing a none-standardized proprietary notation, the industry’s interest vanished [Tsc10]and turned to BPMN. In the already mentioned survey [HW11], only 8% of respondentsacknowledged the use of EPC as a business process modeling language.Among researchers we have seen some interest in EPC, mainly regarding empirical val-idation of EPC process models [Men07] and the supplementing of EPC structural limi-tations, through a more rigorous execution semantics and verification methods [vdA99,
14http://case-tools.org/uml.html [accessed in Feb. 10, 2013].
30
2. PROCESS MODELING 2.5. Process Modeling Languages
Figure 2.4: Event-Driven Process Chain metamodel
HOS06]. (Rating: Scarce.)3. Target Users. EPC, due to its graphical notation and semiformal nature, allows busi-ness analysts benefit some degree of freedom when expressing business processes. Sinceevent-driven process chains are primarily used to model business processes at higherlevel, and humans have non-local knowledge, it becomes more easy to interpret EPCconstructs (e.g. the semantics of the or join). However, when it comes to business pro-cesses implementation, systems engineers can have to rework models in order to removeambiguity and make explicit what was intended by the business analysts. (Rating: Spe-cific.)4. Depth in Verification. EPC provides an informal notation for representing businessprocesses and their environment [Wes07]. Only the core set of EPC elements are formal-ized, and documented. The extended elements of EPC are neither formalized nor welldocumented [Tsc10]. One particular issue is EPC requirement of non-local semantics[Kin03]. For instance in the case of the control flow construct or join, due to the non-localsemantics, the decision on when to use the join needs non-local knowledge [Wes07].Several scientific articles (e.g. [vdA99, HOS06]) are devoted to providing well-definedexecution semantics to EPCs. In [vdA99] EPC models are formalized by mapping toPetri/Nets. Workflow Nets15 [vdA96], an extension of Petri/Net, have been used fortranslating EPC and to facilitate the verification of model’s properties [VdA97] with re-spect to deadlocks and proper termination [Wes07]. (Rating: Semi–formal)5. Kind of Notation. EPC has a diagrammatic notation that allows modeling from abusiness perspective. In a EPC model, entering into a business-relevant state is depictedthrough an event. Events (passive elements that do not provide decisions) trigger func-tions and vice–versa. Functions are fine-grained units of work. Functions, as active el-ements, transform inputs into output. Both inputs and outputs can be data or product-s/services. Functions can include decisions, through associated connector nodes, that
15http://woped.ba-karlsruhe.de/index.php?id=7 [accessed in Feb. 10, 2013].
31
2. PROCESS MODELING 2.5. Process Modeling Languages
impact the behavior of the business process. Connectors model causal ordering relations,i.e. they represent the process logic through splits and/or joins. The bipartite structureof EPC models, in which events and functions alternate, can lead to complex processrepresentations difficult to understand and cope with [Wes07]. (Rating: Diagrammatic)6. Tools Availability. The EPC was mainly disseminated by the modeling tool ARIS(Architecture of Integrated Information System)16, and as a notation used by the SAPR/3 reference model17. EPC is supported by only a few modeling tools of other vendors18
[All10]. The EPC metamodel and process data can be transformed to equivalent XML likerepresentations, such as the EPML (EPC Markup Language) [MN05]. (Rating: Few)
2.5.2.4 Petri Nets (P/N)
1. Extent of Concepts. The Place and Transition Nets or simply Petri Nets (P/N) [Pet62,DR98] is a special kind of graph technique, aimed at representing the behavior of dy-namic systems [Mur89], and being used for business process modeling. The P/N com-prises a small set of modeling elements (Figure 2.5): two kinds of nodes and one connec-tor.The classical P/N, is a bipartite graph composed by a set of places linked via directedweighted arcs to a set of transitions. Connections between nodes of the same type arenot allowed. The places that are upstream from a transition are called its input places.The places that are downstream are the transition’s output places. A transition is said tobe enabled if each input place contains at least as many tokens as the weight of the arclinking it to the transition [MTJ+10]. In a P/N, the static structure of a dynamic systemis represented by a graph. The dynamic behavior is captured by tokens moving betweenthe graph nodes, constrained by firing rules. P/N is a state-based method, i.e., an in-stantiated P/N explicitly represents the state of a process model [Tsc10]. The dynamicbehavior of a system is represented by the consecutive state changes [Wes07].It have been argued P/N lakes expressiveness to deal with some control flow patterns[AH10] namely, the multiple instances pattern, or split, or join and the discriminator pat-tern [Wes07].In the context of business process modeling, P/N have been used to perform models’formal analysis and verification [VdA94, VdAVH96]. With P/N it is possible before theimplementation phase, to simulate and analyze the properties of different alternatives forthe design of a business process. However, the transformation from formal specificationsinto executable processes is not straightforward [VdAVH96]. (Rating: Narrow)
16http://www.softwareag.com/corporate/products/new_releases/aris9/overview/default.asp [accessed in Jun. 11, 2013]
17http://goo.gl/xRfFqv [accessed in Jun. 11, 2013]18In http://en.wikipedia.org/wiki/Event-driven process chain [accessed in Feb. 10, 2013], less than 10
2. PROCESS MODELING 2.5. Process Modeling Languages
Figure 2.5: Classical Petri Nets metamodel
2. Level of Adoption. The classical Petri/Net technique has been used in many appli-cations in several domains, as a formal verification tool (e.g. systems concerning pro-duction, logistic, manufacturing, administrative, workflow, computer systems, and real-time, as well as communication protocols) [VdA94, VdAVH96], given its suitability torepresent reasonable complex behaviors. As business process modeling technique, P/Nhas not achieved a wide acceptance by business analysts and domain experts, due to itsformalism and semantic gap between problem and solution domains. P/N is semanti-cally less expressive when compared to other BPMLs. The adoption of BPMN, AD orEPC, overcame the P/N usage. (Rating: Scarce)3. Target Users. As was previously mentioned, P/N main applications have been in de-scribing and studying systems that are characterized as being concurrent, asynchronousdistributed, parallel, nondeterministic, and/or stochastic. Therefore, P/N usage has beenconfined to people with formal methods background. P/N relevance is recognized whensoundness evidence is required, either by IT specialists or computer science researchers.When describing real cases, P/N is not concise and flexible enough for modeling high-level and complex business processes [LA94]. P/N usually generates large models, hardto understand and manage. Therefore, it was consensually agreed that P/N was not themost appropriate tool to be adopted by business engineers, more concerned with theproblem domain [MTJ+10]. Using P/N for business process modeling leads to the lostof structural and semantical information [Tsc10], since the functional perspective is onlyconveyed abstractly [Wes07]. (Rating: Specific)4. Depth in Verification. P/N allow for many analysis techniques. These techniques canalso be used to analyze modeled business procedures. P/N theory, enables, for instance,the verification of business processes’ safety properties (e.g. reachability, boundedness,liveness, and reversibility and home state), correctness (e.g. absence of deadlocks andtraps), and invariance properties (e.g. by proving that, for an external observer, twoprocesses behave logically the same way, which allows to evaluate alternative designs),as well as business processes’ performance measures (e.g. response time, delay time,
33
2. PROCESS MODELING 2.5. Process Modeling Languages
occupation rate) [VdAVH96].In spite of P/N constructs being semantically far from the abstractions that business pro-cess modeling users are familiarized, the P/N representation serves as a solver-independentmedium between the problem domain and the analysis techniques [VdA94] for morewell-known BPMLs such as BPMN [DDO07], AD [Fah08] or EPC [vdA99]. So, P/Nformalism can be seen as a low-level language, almost as a kind of assembly languagelevel modeling language, i.e. a formal supplement of other business process languages[MTJ+10]. (Rating: Formal)5. Kind of Notation. P/N is a graphical representation of systems with mathematicalfoundation, aiming at assisting analysis of the structure and dynamic behavior of mod-eled systems, especially concurrent ones [Pet77].As a graphical tool, P/N can be used, in certain contexts, as an instrument of commu-nication for requirements gathering, modeling and analysis [Gia01]. Places can be usedto model the flow of work, whereas transitions can be mapped into activities. A tokencan represent organizational resources or entities, such as documents, materials, services,equipments, or persons. In addition, tokens can be used in these nets to simulate the dy-namic and concurrent activities of systems [Gia01]. The transitions linked together byplaces define a partial ordering of tasks. Transitions can also be added to the model torepresent control flow elements (such as fork and join), which provide routing, paral-lelization, or synchronization of the work flow.Classical Petri/Nets describing actual systems, often distributed and temporal constrained,tend to be complex and extremely large. To solve this problem, the classical Petri/Netmodel gained extensions such as color, time, and hierarchy: ’colored’ tokens facilitate themodeling of objects’ attributes; the ’time’ construct allows to model the temporal be-havior of a system; and the ’hierarchy’ construct enables the decomposition of complexsystems. A number of extensions to the basic P/N formalism were proposed to overcomethe P/N initial limitations, such as High-level Petri/Nets [VdA94], GSPN (General-ized Stochastic Petri/Nets) [Mar90, CMBC93], CPN (Colored Petri/Nets) [Jen94, Jen97,Jen98, KCJ98], OOPN (Object-Oriented Petri/Nets) [Lak95], and TPN (Time Petri/Nets)[LS00, dFS08].A precise and unambiguous description of the business process behavior is the result of amodeling process using P/N. Such preciseness and the firm mathematical foundation ofP/N have resulted in the affluence of analysis methods and tools [VdAVH96]. (Rating:Diagrammatic & Mathematical)6. Tools Availability. The richness of P/N analysis techniques have made bloom theavailability of tools19. More than fifty options are available for analyzing classical P/N,and dozens for more advanced P/N, such as High-level P/N, P/N with Time, HybridP/N, Stochastic P/N, Object-oriented P/N, Modular high-level P/N, Colored PN, etc.(Rating: Many)
19http://www.informatik.uni-hamburg.de/TGI/PetriNets/tools/quick.html [accessed in Feb. 10, 2013]
34
2. PROCESS MODELING 2.5. Process Modeling Languages
2.5.2.5 Yet Another Workflow Language (YAWL)
1. Extent of Concepts. Process modeling constructs in YAWL encompasses the con-ceptual and the execution level [Jur11]. The motivation for the development of YAWL[tHvdAAR09] was the ability to support directly all workflow (control flow) patterns20
[Wes07, AH10]. The basis for the semantical foundation of YAWL was the P/N and theirvarious extensions, in particular Workflow Nets21 and reset nets [LS07]. Therefore, thesemantics of YAWL has been defined in terms of a large Colored P/N.YAWL uses extended workflow nets as building blocks, enhancing them with notationalconvenience (e.g. direct arcs between transitions), explicit split and join behavior thatcan be attached to transitions, nonlocal behavior (on the firing of a transition, parts of theextended workflow net are cleansed of tokens), and the handling of multiple instancestasks.Conditions are represented by circles and tasks are depicted by rectangles (see in Figure 2.6the main constructs of YAWL). The initial condition and the final condition of an extendedworkflow net are labeled with specific symbols. There are several notational extensionsfor tasks (composite tasks mapped to an extended workflow net, tasks with multipleinstances, composite tasks with multiple instances). The graphical representation of thesplit and join behavior of a task is equivalent to that in workflow nets. For each task acancellation region can be defined.Since P/N already supported most of the control-flow patterns, the YAWL’s designerstook the P/N as a starting point to extended the P/N formalism with three main con-structs, namely or-join, cancellation sets, and multi-instance activities. These three conceptsaimed at supporting the patterns that were not directly supported in P/N, namely syn-chronizing merge, discriminator, M-out-of-N join, multiple instance with no a priori runtimeknowledge and cancel case. In addition, YAWL added other syntactical elements to P/Nin order that the language could capture other workflow patterns, namely simple choice(xor-split), simple merge (xor-join), and multiple choice (or-split).It was pointed out not completely satisfactory YAWL representation of some advancedcontrol flow patterns [Wes07], including the discriminator and the M-out-of-N join. Onthe other hand, the graphical representation might become cumbersome if the tasks andconditions are spread across a large workflow specification and if multiple tasks removetokens. (Rating: Broad)2. Level of Adoption. YAWL is, until now, mainly a research project with scarce adop-tion by industry22. A reason for which YAWL did not catch the attention of the processmodeling community, could be its academic origin, as well as the strong competition ofalready existent BPMLs, such as BPMN, AD and EPC.According to Börger, YAWL fails to provide the practitioners with suitable concepts to
20http://www.workflowpatterns.com/21http://woped.ba-karlsruhe.de/index.php?id=7 [accessed in Feb. 10, 2013]22In http://www.yaug.org/projects [accessed in Jun. 14, 2013] is only referred one YAWL implementation
in a clinical environment.
35
2. PROCESS MODELING 2.5. Process Modeling Languages
Figure 2.6: Yet Another Workflow Language metamodel
guarantee that business processes are best described for the purposes of analysis and subsequentautomation [B12]. Furthermore, workflow patterns, which is the basis for YAWL expres-siveness, seems not being perceived, by process modelers in general, as the fundamentalaspect for being the basis for the construction of a business process modeling language.The YAWL notation itself is a drawback since it is only familiar to people closely relatedto workflow nets and P/N notations. (Rating: Scarce)3. Target Users. The graphical nature of YAWL allows for models used from the concep-tual stage to the execution level, hence covering different levels of abstraction.Given the similarities between logistics and administrative processes, both aiming at a re-duction of throughput times and resources, it is advocated the advantage of using YAWLas a modeling tool in both kind of industrial systems [VdA94].The YAWL language targets both business analysts and process implementers, both inindustry and academia. (Rating: All)4. Depth in Verification. YAWL has formal semantics, thus leading to precise and un-ambiguous descriptions. The behavior of the modeled systems implemented in YAWLcan be analyzed simillary to those expressed with P/N.The overall idea for expressing the execution semantics is that each task is represented byan individual state transition diagram. A state transition diagram of a task specifies itscurrent state. The state of the process instance is then represented by the combined stateof all tasks involved in the process instance, plus conditions that are currently met at theprocess level.During the design of the language, it was concluded that some of the extensions thatwere added to P/N could not be re-encode back into plain P/N. Hence, the originalformal semantics of YAWL was defined as a labelled transition system and not in terms ofP/N.The fact that YAWL is based on a formal semantics has enabled the implementation ofseveral techniques for analyzing processes in the YAWL engine. (Rating: Formal)
36
2. PROCESS MODELING 2.5. Process Modeling Languages
5. Kind of Notation. The notational elements of YAWL borrows most constructs ofworkflow nets. Process models use extended workflow nets as building blocks, and mul-tiple extended workflow nets involved in a process model specification can be connectedto each other by composite tasks. As a high-level P/N, YAWL combines a rigorous math-ematical notation with the graphical notation.There are differences with respect to handling tokens. Conversely to P/N, in extendedworkflow nets, tokens reside at transitions while the transition is being executed. Thisalso means that the execution of a transition takes time, another difference to P/N and toworkflow nets, in which the firing of transitions is not time consuming [Wes07]. (Rating:Diagrammatic & Mathematical)6. Tools Availability. YAWL currently was only implemented by one tool23, which al-lows the specification and execution of workflow models. In [DDDGB08] a tool is de-scribed that carries out the transformation between BPMN diagrams to YAWL nets. (Rat-ing: Few)
2.5.2.6 Subject-oriented Business Process Management (S-BPM)
1. Extent of Concepts. S-BPM is a communication view based on subjects, which cancompose a business process orchestration or a choreography. The S-BPM modeling paradigmcomprises a small set of modeling elements (Figure 2.7). The constructs used to modelany process allows direct transformation into executable form. Each business process con-sists of two or more subjects that exchange messages. Each subject has an internal behav-ior (control flow between different states), which are receive and send messages and dosomething. There are more elements used, but they are not necessary, and only exist forsyntactical sugaring.In S-BPM the focus is put on the acting elements within a process: the subjects. Theyexecute and synchronize their activities by exchanging messages. A message is based onthe structure of sentences in natural languages (subject, predicate and object). The subjectis the initiator of an activity, the activity is the predicate and the target of the activity theobject of the sentence [Fle10].The sequence in which subjects carry out their activities is described in the subject behav-ior. Each subject has an input pool in which the sending subject deposits the messagesfor this subject. The corresponding subject accepts messages by removing them from theinput pool. By configuring the input pool of a subject we can specify which messagesfrom which subject are received synchronously or asynchronously. Messages transportbusiness objects which contain information that can be hierarchically structured.In a send state, a subject tries to send several messages alternatively. In a receive state asubject can accept several messages alternatively. Internal actions of subjects are definedon internal data like business objects. Internal actions can change business objects orcheck the values of elements in a business object, or both.
23http://www.yawlfoundation.org/ [accessed in Feb. 10, 2013]
37
2. PROCESS MODELING 2.5. Process Modeling Languages
Processes are connected via the communication of subjects in the connected processes.Subjects in one process exchange messages with subjects in other processes they are con-nected to. Service processes are very similar to connected processes. In a service processthere is a subject which is visible to all other processes. This subject is called the interfacesubject. This interface subject accepts messages from subjects in any other process. Multi-processes are an extension of connected processes. If a multi-process receives a certainmessage from a connected process, a copy of the multi-process is generated.A process can be connected to another process. This connected process can in turn be con-nected with other processes and so forth. In this way, hierarchical networks of connectedprocesses can be constructed. Each process can consist of subjects and other processes.There can be any number of internal subjects not visible to other processes.If it is not required that specific actions being executed in a certain order, a choice operatorcan be used to express the overlapping execution of action sequences. Normally, eachsubject has only one active state. An active state is the state representing the next actionto be executed by the corresponding subject.Identical activity sequences are used in behavior specifications of a subject at severalplaces in the behavior description or in several different subjects. These sequences canbe defined once as a macro. Macros can be embedded in subject behaviors where thecorresponding macro behavior is required.Since S-BPM is based on a process specification language, with a formal semantics, itallows that executable workflows can be generated automatically from a process model[Fle10]. The search for new methods such as S-BPM has been motivated by the demandto better support human collaboration and communication in business processes (e.g.ad-hoc processes, empowerment, human interaction workflows), which seems to be notwell supported by current approaches. (Rating: Broad)
Figure 2.7: Subject-oriented Business Process Management metamodel
2. Level of Adoption. S-BPM is still a language in development. Its adoption (only adozen of reported installations24) seems mostly to occur in research projects instead of
24http://www.metasonic.de/en [accessed in Feb. 10, 2013]
2. PROCESS MODELING 2.5. Process Modeling Languages
full fledged BPMS. Originally launched as an academic project, S-BPM has slowly beenacknowledged by the business process modeling community, as an alternative perspec-tive for modeling business processes in organizations [FSS+12a]. (Rating: Scarce)3. Target Users. Process modelers that participate in organizational developments usenatural language constructs and e-mail-like communication patterns between them, whendescribing business processes. Hence, all stakeholders of an organization can contributeto the process specifications, and the delivered specifications processed without trans-formations. In subject-oriented modeling scheme actors are acknowledged as the entrypoint for modeling, leading to an approach that takes into account standard sentence se-mantics (subject, predicate, object). By using subjects, stakeholders intends to avoid thatconveyed information be reduced either to content or functional business logic. (Rating:All)4. Depth in Verification. The language of S-BPM is called Parallel Activity SpecificationSchema (PASS). It is based on the grammar of natural languages, Calculus of Communi-cation Systems (CCS) theory [Mil99] and the concepts of object-oriented programming.These concepts are extended by additional pragmatic elements, which allow a specifi-cation of various recurring behavioral aspects in business processes. It were also addedadditional elements to support the structuring of highly complex and extremely largeprocess systems [FSS+12b].CCS is a process algebra [Mil99] used for algebraic modeling of parallel processes andconsists of elementary actions and operators for joining actions. The main objective ofCCS was to provide a mathematical framework to describe communicating systems ina formal way. Processes can interact with the neighbors or independently perform ac-tivities in parallel. The aim of CCS is to model the communication between processes,e.g., to investigate their equivalence. A process uses ports as enablers of communica-tion with other processes, whereby each port has a name. The active element in CCS,the actor, is seen as essential, while predicate and object play a subordinate role. Thus,CCS can be considered a subject-oriented method. Subject orientation include actors intothe structures of coarse-grained business processes providing directions to their decom-position for implementation at a detailed service level. Once the patterns regarding theinteraction among actors (subjects) has been refined regarding the exchange of messages,the program code can be then automatically generated in a more suitable way. Execut-ing process models provides immediate organizational user experience, since it abstractsfrom implementation details related to programming languages or software execution.(Rating: Formal)5. Kind of Notation. The information in S-BPM is presented in diagrammatic form,by focusing on the interaction among subjects. The principles of decomposition andhierarchy-specific associations is also followed by S-BPM. The subject, predicate and ob-ject of concern have, in conceptual terms, the same importance for humans when realityis modeled. The use of identical formal representations drives to concise models in termsof flow control. S-BPM triggers the shift of process modeling paradigms from the focus
39
2. PROCESS MODELING 2.5. Process Modeling Languages
on functions or activities to the emphasis on business entities and role-specific entities(actors, subjects).As mentioned, the S-BPM methodology is anchored in the mathematical framework ofCCS-Calculus. Milner states that every interesting concurrent system is built from inde-pendent agents which communicate in a synchronized way [Mil99]. So the objects whosebehaviors are modeled are called agents. An agent can be seen as a term for a locus ofactivity, a process, or a computational unit. The agent’s behavior is defined by the actionit can perform and is represented using algebraic expressions. The notion of agent is CCSis mapped to the notion of subject in S-BPM. (Rating: Diagrammatic & Mathematical)6. Tools Availability. S-BPM currently was only implemented by one tool (MetasonicSuite25), which allows the specification and execution of workflow models. On the otherhand, I2PM26 serves as a community platform to bundle research and development ac-tivities in the field of S-BPM. (Rating: Few)
2.5.2.7 Business Process Execution Language
1. Extent of Concepts. BPEL is a industry standard [OAS07] that defines a model anda grammar for business processes’ orchestration and collaboration. BPEL ensures themeans for connecting web services, as well as specifying how web services can be jointlyused to provide more complex functionality implementing business processes. BPELintends to give high flexibility to business processes execution and can be regarded asa foundation of the technical architecture of workflow oriented web applications [ZJ08].Figure 2.8 depicts the main constructs of BPEL.BPEL introduces mechanisms for dealing with business exceptions and process faults.Moreover, BPEL defines how activities within a unit of work are to be compensated incases where exceptions occur or a participant requests the operation’s reversion [CT12].Broadly speaking, one can considerer two categories of activities in BPEL: basic activitiesand structured activities. Basic activities are contained in structured activities, and in-clude constructs such as Invoke, Receive, Reply, Assign, Throw, Wait, Empty, and Terminate.Invoke is used to invoke web service operations, Receive and Reply are used to provideweb services operations. On the other hand, structured activities include, for instance, theconstructs Sequence, Switch, While, Pick, Flow. Pick allows to block and wait for a suitablemessage to arrives or for a time-out alarm to go off. Flow allows to specify one or moreactivities to be performed concurrently.While a powerful language, BPEL is difficult to use. Its XML representation is very ver-bose and requires experienced users. It offers many constructs and typically things can beimplemented in many ways (e.g. using links and the flow construct or using sequencesand switches). As a result, only experienced users are able to select the right construct[vdABL08].BPEL shows the trend towards process-oriented programming. One of the aims of BPEL
25http://www.metasonic.de/en [accessed in Feb. 10, 2013]26http://www.i2pm.net/open-s-bpm [accessed in Feb. 10, 2013]
2. PROCESS MODELING 2.5. Process Modeling Languages
is to function as a serialization format and execution language for BPMN diagrams [CT12].Since BPEL, as a process language is not complete, it is often used in conjunction withother languages to fill in the existing gaps. In addition, BPEL is often tied to proprietaryimplementations of workflow or integration broker engines.It was also acknowledged BPEL failure in supporting human tasks. Those tasks are usu-ally allocated to human actors and require from them that other actions be completed,possibly involving physical resources. Some engines already provide extensions to BPELfor human tasks. It is expected that those extensions could be standardized in futureBPEL versions. Last version of BPEL supports several of the Workflow Patterns [RM06].As previously mentioned, BPEL is dedicated to process execution, and is designed tomeet the needs in the environment of web services. Hence, BPEL does not include con-structs amenable to be used for modeling business processes at an higher level of abstrac-tion. (Rating: Narrow)
41
2. PROCESS MODELING 2.5. Process Modeling Languages
2. PROCESS MODELING 2.5. Process Modeling Languages
2. Level of Adoption. In the already mentioned survey [HW11], 6% of respondentsacknowledged the use of BPEL as a business process language. This figure is well belowthe one got by BPMN (72%), but above any other process execution language (e.g. YAWL,P/N).BPEL has yet many limitations that must be considered when searching for the finalchoice for serialization of BPMN diagrams. The transformation between these two lan-guages is not straightforward, and has generated some debate among researchers [Gao06,WDGW08]. Indeed, there is still no complete solution that would fully cover the trans-lation of any kind of BPMN diagram to BPEL. Current techniques focus either on somesubset of BPMN, most often the basic subset, or some specific BPMN elements. Cur-rent transformations also impose constraints on the structure of BPMN diagrams [Jur11].Many BPMN graphical elements cannot be represented in BPEL and many attributes andproperties of BPMN constructs, lack of a precise meaning and have no representation inBPEL [CT12].BPEL has attracted the attention both from practitioners and the researchers. The latterhave been working in defining mapping rules, and formal verification of BPEL [Jur11]. Itis expected that the adoption of BPEL will increase with the diffusion of BPMS engines,as well as due to mitigation of limitations in the current version. (Rating: Moderate)3. Target Users. BPEL is not used by business analysts in analysis and design phases,since it offers no standardized graphic notation. Hence, BPEL language is not suitable forbusiness analysts’ tasks such as value chain analysis or processes’ optimization analysis[Lon04].Standing on top of the web service specification stack, BPEL became the de facto standardof process execution [Lon04]. Therefore, BPEL is mostly used by process implementers.(Rating: Specific)4. Depth in Verification. BPEL is essentially a service modeling language based onblocks. BPMN process modeling language was established to be the main upstreamprovider of specifications to BPEL. As aforementioned, there is a mismatch betweenBPMN (a process model language) and BPEL (a service model language) [Jur11]. Thedifferent languages’ paradigm, together with the incomplete BPMN standard definitionshinders the transformation between them, making sometimes even impossible the veri-fication of process implementation based on process modeling.The threads on research aiming to fill the gap between BPMN and BPEL are mainlybased upon formal approaches. They include for instance: recognizing patterns in BPMNfor defining standardized mapping rules; formal verification of diagrams; conversionto Petri nets, with existent methods of verification, and their subsequent conversion toBPEL; and formal logic approach using functional logic programming languages. Re-searchers have already captured the formal semantics of subsets of BPEL using formalisms,such as petri nets, process algebra and finite state machine. This has contributed for thedevelopment of static analysis tools for BPEL. [Jur11]. (Rating: Formal)5. Kind of Notation. BPEL is a notation for specifying business process behavior based
43
2. PROCESS MODELING 2.5. Process Modeling Languages
on web services, and is layered on top of Web Services Description Language (WSDL),XML Schema, and XPath. WSDL messages and XML Schema type definitions providethe data model used by BPEL processes. XPath provides support for data manipulation.The external resources and participants are represented as WSDL services. Graphicalinformation for BPEL is optional and tool dependent. (Rating: Mathematical)6. Tools Availability. Since BPEL is driven by a standardization committee with severalIT industry players [Jur11], it is supported by a relevant number of tools27 (both propri-etary and open-source).BPEL as a business process implementation language based on Web services is executedby BPMS engines. Some platforms besides supporting the execution of BPEL code, pro-vide also the ability to graphically specify BPEL processes in a proprietary notation.However, these tools are mostly focused on BPEL as a process execution language anddo not provide enough level of integration with earlier stages of process modeling, suchas with BPMN [Jur11]. (Rating: Some)
2.5.3 Assessment of BPMLs
In Table 2.1, we summarize the BPMLs’ assessment based upon the previously definedtaxonomy. YAWL and S-BPM, seems to be the most well grounded process modelinglanguages, however without the level of adoption and tools availability of other lan-guages. In an overall appreciation BPMN seems to be nowadays the most well positionedBPML. The main weaknesses identified in BPMN regards models verification (supportedby techniques and properties checkers in other languages) and the notation (based upona metamodel that enables models’ syntactical verification but lacking the semantic formaldefinition).
Table 2.1: Summary of BPMLs Assessment
BPMN AD EPC P/N YAWL S-BPM BPEL
Extent of Concepts O O OLevel of Adoption O O O O Target Users O O ODepth in Verification Kind of Notation O O O Tools Availability O O O
Legend: assigned values in each dimension converted to symbols (i) O (ii) (iii)
Based in the previous assessment, we have chosen BPMN in this dissertation, as theprocess modeling language, to address our research problem of quality in process mod-eling. In brief, the best characteristics of BPMN are:• BPMN is currently the business process notation most used among process model-
ing practitioners [HW11], with a preference rate above 70%;
• BPMN is a process modeling standard backed up by OMG, so the language defini-tion is based upon a metamodel [BPM11] built with UML, the notation which is thede facto standard for modeling software engineering artifacts [OMG07a, OMG07b];• BPMN is one of the most recent BPMLs, so it is grounded on the experience of ear-
lier BPMLs, which ontologically makes it one of the most complete BPMLs [RIRG05,RRIG09];• BPMN has transformations to other available notations, such as CSP [WG08] and
P/N [DDO07], which allows the use of available tools for formal verification;• BPMN is the language with more modeling tools available.
2.6 Conclusion
The chapter begins by giving a historical context for the emergence of the process paradigm(section 2.1), and by introducing the main concepts regarding business processes and pro-cess modeling (section 2.2). Is put some care, to differentiate processes’ usage in severaldomains (section 2.3). We clarified also the particular perspective, of the multidimen-sional concept of quality, relevant in the context of this dissertation (section 2.4). Theassessment made to current process modeling languages allowed to choose the betterone positioned nowadays (BPMN), for addressing quality upon process modeling (sec-tion 2.5).
45
[This page is intentionally blank]
3Analysis of the BPMN
"The limits of my language means the limits of my world."
Context: Currently, the BPMN is the most used language among process modelers.Objective: Assess the suitability of the BPMN for being used to produce good qualityprocess models.Method: A set of BPMN tools was surveyed, for assessing whether they assist in de-signing good quality process models.Results: The BPMN models produced by the surveyed BPMN tools revealed flaws andpoor quality. The main reason is due to the informal prescription of the rules in theBPMN standard.Limitations: The survey on BPMN tools considered a small set of BPMN rules.Conclusion: The BPMN has limitations that must be overcome in order to be able tobetter address the quality of process models.
47
3. ANALYSIS OF THE BPMN 3.1. Introduction
3.1 Introduction
The purpose of BPMN [BPM11] is to provide a notation understandable by differentkinds of process modelers and users: (1) process analysts that sketch the initial docu-mentation of the processes; (2) process implementers which are responsible for actuallyimplementing processes; (3) business users which are accountable for processes’ usageand monitoring. This intended broad coverage of process models resulted in the usageof BPMN in many applications, from process documentation and improvement scenariosto technical applications of process modeling, such as workflow engineering, simulationor web service composition [IMR09].
The need for accurate specification of process models has emerged as a primary req-uisite of conceptual modeling activities. This need is present in various activities in theorganization, namely when producing processes’ specifications to meet various regula-tory and legal requirements (e.g. SOX [Lan03], BASEL III [Sup10]), for the analysis anddesign and development of process-aware information systems [DVDATH05], service-oriented architectures [Erl07], and web services alike systems [IMR09].
Although BPMN can be used for all these purposes, the business and technical mod-els produced are quite different in nature. The main focus of BPMN business modelsfor documentation purposes is on the comprehension of the basic process flow. Thus,usually only the happy path is depicted, avoiding excessive details [Sil09]. Exception han-dling and abnormal situations are quite often disregarded. On the other hand, executioncapabilities of the BPMN language are of interest when producing technical models. Theneed of process developers is on translating BPMN models into some machine readablelanguage either for models’ sharing across multiple domains and using many differenttechnologies (e.g. XMI, XPDL), simulation, and execution in distributed environment(e.g. integrating BPEL and web services standards).
BPMN as a process modeling language joins different business centric points of view,with different levels of abstraction. Those perspectives are related to the analysis anddesign effort, as well as the simulation and execution of processes. The BPMN standardthus provides the ability for accommodating different sorts of constructs either for staticgraphical representation of models, as well as properties that provide information forsimulation and execution of processes. Currently there are already tools (e.g. Bizagi1)that directly execute BPMN structures without having to transform them into BPEL oranother format first.
To harness all the potential of the BPMN standard, efforts still have to be carried outto tighten the different levels of abstraction of the process life cycle. As we will see,by ensuring the compliance of models made in primary stages of process modeling withBPMN well-formedness rules, it would be enforced the correctness of those process mod-els and therefore their reuse from modeling to enactment and monitoring.
Nevertheless, BPMN fills the set of criteria established [Lon04] as a requirement for
1http://www.bizagi.com/ [accessed in Feb. 10, 2013]
48
3. ANALYSIS OF THE BPMN 3.2. Modeling with BPMN
being a business process modeling standard, namely: (1) a notation widely adopted ei-ther by business and technical people, due to its tool independence [HW11]; (2) a meta-model which provides an abstract syntax, through a consistent vocabulary of conceptsand relationships; (3) the ability through the notation and metamodel of focusing andanalyzing particular details of the process model (e.g. control-flow, data, organization);and (4) provide an exchange format for process models (e.g. XMI, XPDL).
The remain of this chapter is structured as follows: section 3.2 addresses the recentevolution of BPMN, until to become the most widely used process modeling language,as well as its main characteristics. Section 3.3 provides an overview of the BPMN meta-model and introduces the main constructs used in the process orchestration. Some atten-tion is also given to flaws detected in the BPMN metamodel, as well as proposed con-cepts not covered by current BPMN standard. In section 3.4 we highlight the weaknessesof BPMN which impact upon quality of BPMN models. Finally, section 3.5 concludes bysummarizing the chapter.
3.2 Modeling with BPMN
From BPMN 1.2 onwards (Figure 3.1), increased substantially the number of changes tothe BPMN standard. The updates made included either the addition of new constructs orthe change of the elements’ properties defined in the previous versions of the standard,in order to resolve known inconsistencies and ambiguities [CT12].
The BPMN 2.0 specification is a step forward extending the scope and capabilitiesof the business process modeling language. The syntax of the language is formalizedthrough the definition of a meta-model representing the language’s constructs and theirrelationships. Since BPMN [BPM11] is one of the most recent process modeling lan-guages, it is ontologically one of the most complete which is available [RIRG05, RRIG09].While, for instance, a UML Activity Diagram can be built from around 20 different mod-eling constructs [OMG07b], an orchestration BPMN process model has almost 100 differ-ent modeling constructs (see Figure 3.2), including among others 5 types of subprocesses,9 task types, 6 gateway types, 3 activity markers, 5 data types, 3 sequence flow types, and51 event types (see Figure 3.3). This gave to language even more expressiveness for dia-gramming, as well as for execution purposes [CT12].
The strong argument for BPMN, as a process modeling language, came since its in-ception with BPMN 1.0. According to an assessment made by Stephen White [Whi04],the standard is suitable for representing almost all aspects of the control-flow of the Work-flow Patterns [AH10]. Control-flow patterns are used as a general pattern to compare theexpressiveness of process modeling languages, since they are independent of any specificprocess modeling languages [Wes07]. Control flow patterns are defined at the processmodel level (with execution semantics applied to process instances). Some examples ofcontrol flow patterns include sequence, and split, and and join, as well as exclusive or splitand exclusive or join.
The semantics of many of the BPMN constructs came clearly from the field of exe-cutable process specifications. Among others, one can refer special constructs dealingwith loops, exception handling, and transactions. Programmers and IT-specialists aremore familiar with these sort of elements. Process analysts typical only use a restrictedpart of the whole notation (see [zMR08]) and normally omit items with more complexsemantics. However, some BPMN experts [Sil09] consider that these technical constructsshould also be used by process analysts, in order that BPMN models are able to showbusiness-relevant exceptions and their handling. In favor to this argument they recall thefamiliar 80-20 rule, which says that 80% of the costs, delays, and errors come from 20%of the cases - the ones that are the exceptions to the happy path. So, process analysts, asdomain experts, should be the ones accountable for modeling exceptions.
The large current number of diagrammatic symbols of BPMN increased their per-ceived complexity [ZMRI07]. As acknowledged in [IMR09], the complexity of processmodeling languages affects the ability of process modelers to model the domain usinga process-centric approach. In [May89] it is claimed that the learning performance, de-pends on the person’s characteristics, as well as by learning materials and the way con-cepts are presented. Language complexity is thus a significant issue, because it can affectthe learnability, the ease of use and overall diffusion of a language.
Recent studies [RIRG06, MR08, ZMRI07] on process modeling languages indicatedalso that the perceived complexity affects negatively the usage of those languages. How-ever, paradoxically, also showed that languages such as BPMN, with a larger vocabulary,are used more frequently than others with a more restricted vocabulary (e.g. P/N, AD).
50
3. ANALYSIS OF THE BPMN 3.2. Modeling with BPMN
Figure 3.2: Concrete syntax of BPMN
51
3. ANALYSIS OF THE BPMN 3.2. Modeling with BPMN
Figure 3.3: Event types in BPMN (Source [BPM11])
52
3. ANALYSIS OF THE BPMN 3.2. Modeling with BPMN
There are several approaches, from different fields, which have been used in studies(e.g. comparison and analysis) about modeling languages in general, and business pro-cess modeling languages in particular. Some of those fields are: semiotics [KLS95, KS00,Kro03, Sin06], ontology [DGMR03, RGI04, RIRG05, GRI05, RI07, RRK07, RIG07, RRGI09,RRIG09], measures [BBM96, BRvU00], empirical studies [SRD03, GL06, OHNAEE+07],meta-modeling [LS97, RM98, zM99, RG02, DGMR03, BV10], and theoretical foundedstudies [HR85, vdA98, LS07, BBKK04].
We highlight here the ontology perspective, given its broad usage. Ontology-basedevaluations use as basis a model of real-world concepts (a representational ontology)and involve a mapping between the model and the constructs of the modeling language.The Bunge-Wand-Weber ontology (BWW) [WW90a, WW90b, STW03], has been widelyused for assessing business process modeling languages2. BWW allows the evaluationof the degree of coverage by language constructs of certain dimensions (e.g. structural,dynamics, components) regarding process models.
BWW was used in a study to investigate the extent in which different subsets ofBPMN vary in their complexity [RIRG06]. The findings show that BPMN exhibits a largeamount of language complexity due to constructs and constraints that are not graphicallyrendered. This indicates that the underlying rules and constraints of the language are asignificant source of complexity. Those findings lead also to two relevant conclusions:
1. users tend to reduce the complexity of the language by ignoring the language’sunderlying rules, when making practical usage of it. In the particular case of BPMNmodels, this means that process modelers violate BPMN underlying grammaticalrules, using constructs outside of their designated scope;
2. not all BPMN constructs are equally used for modeling purposes. The bulk of thehidden complexity comes from constructs that are seldom used in practice. Thus,most of the complexity of the language is invisible to process modelers since theydo not use the constructs whose semantic is more difficult to apprehend;
In spite of a design far from being elegant, BPMN has found widespread acceptanceby industry and a wide-ranging tool support. This suggests the importance given tothe requirement of BPMN expressiveness even at the expense of its increased complex-ity. Researchers have shown that BPMN suffers from a number of shortcomings, beingcomplexity one of them, which impacts the clarity of the language although not affectingthe language’s acceptance [IMR09]. The findings also indicate that BPMN users, con-sciously or not, take active steps to reduce language’s complexity. Anecdotal evidencesuggests that a common way of such reduction is through development and enforcementof modeling conventions within organizations. Modeling conventions typically mitigatethe complexity of BPMN through a selection of constructs with an overall lower com-plexity and defining the patterns for better usage of those constructs.
Summing up, if BPMN modelers are given the freedom to combine the large plethora
2Green and Rosemann counted more than 25 papers that applied the BWW ontology for the analysis ofmodeling grammars [GR05].
53
3. ANALYSIS OF THE BPMN 3.3. BPMN Metamodel
of modeling constructs available, in the absence of any verification / recommendationfacility embedded in the used modeling tool, inconsistent and/or even invalid modelscan easily be produced. Incidentally this is not different from the problem that occurswith some other languages (e.g. EPC, Workflow Nets) [CMNR06]. The great flexibilityof BPMN allows that non savvy modelers might combine BPMN elements in a faultyand unstructured process model. Moreover, it is not straightforward that examining agraphical description of a process could allow users to easily discover inconsistencies orwell-formedness errors (e.g. split not matching a corresponding join). However, userscan impose themselves restrictions in the usage of the BPMN language, thus deliberatelylowering the level of the language’s complexity. The introduction of restrictive modelingconventions, i.e., modeling best-practices, makes the constructs easier to apply and theresulting models easier to communicate.
3.3 BPMN Metamodel
The BPMN 2.0 standard, was a step forward in the alignment of the originally BusinessProcess Management Initiative3 (BPMI) process modeling notation with OMG’s initiativeof Model-Driven Architecture (MDA) [OMG03b]. The BPMN language definition isbased upon a metamodel built with the UML [OMG07a, OMG07b], the standard de factofor software engineering modeling.
The specification defines different types of conformance that BPMN tool implementerscan adhere to, namely regarding:• process modeling – elements that are part of the orchestration in a single process,
as well as elements that participate in the collaboration among processes. A collab-oration is the synchronized interaction of two or more processes without a centralcontrol. Processes communicate by exchanging messages. Collaboration diagramsare used for documenting the co-operation among several organizations [All10].• BPMN process execution – the interpretation of the Activity life-cycle;• BPEL process execution – mapping of a BPMN model to WS-BPEL; and• choreography modeling – a set of elements that puts modeling emphasis in the in-
teraction among participants. This includes choreography and conversation diagrams.The metamodel also has got additional language constructs that cannot be repre-
sented in diagrams. Such constructs are required, for example, by process engines tocapture the necessary additional information for process execution. BPMN 2.0 definesalso an extensibility mechanism for both process model extensions and graphical exten-sions, refines event composition and correlation, and extends the definition of humaninteractions [SWB+12].
The BPMN standard specification [BPM11] can be referred, for the definition andmeaning of each element, as well as for the rules about how they can be connected and forthe connections meaning. However, it is a too complex technical document to be suitable
3http://www.bpmi.org/ [accessed in Feb. 10, 2013]
54
3. ANALYSIS OF THE BPMN 3.3. BPMN Metamodel
to normal business modelers. Besides, the standard does not provide guidance on howthe modeling notation should be used to attain a comprehensible and expressive BPMNmodel. Moreover, a great deal of definitions and rules are only informally presented inplain English.
A metamodel (M2 level according to the MDA paradigm) describes the abstract syn-tax of a language by means of meta-classes, meta-associations and cardinality constraints.
The BPMN metamodel includes elements from three diagrams, targeting the follow-ing different purposes:
1. for modeling processes’ orchestration and collaboration diagrams;2. to simplify the perspective of collaboration diagrams through conversation diagrams;3. for modeling participant’s interactions through the choreography perspective.
In this dissertation, from the full metamodel that includes 151 meta-classes and 200meta-associations, we only consider the subset of elements concerning the orchestrationand collaboration diagrams (depicted in Figure 3.4 and explained in more detail in section3.3.1). This is, by far, the most well-known and used by practitioners subset of BPMN,since it was already present in version 1.
In the next section we will introduce the main concepts and connections of the or-chestration perspective, as described in OMG’s BPMN metamodel. A symbol relates theelements of BPMN concrete syntax depicted in Figure 3.2, with the correspondent meta-class in the abstract syntax representation (Figures 3.4 and Figure 3.5), mapping by thisway, the M2 and M1 layers of the MDA for BPMN.
3.3.1 Detail of BPMN Metamodel
In this section we depict the diagrams with the meta-classes and meta-associations usedin BPMN for process orchestration. For the sake of understandability of the diagramswe labeled the meta-classes with a symbol. This allows to map the meta-classes convey-ing the abstract syntax, with the correspondent graphical element in the concrete syntaxdepicted in Figure 3.2.
The metaclass Process (Figure 3.5) describes a sequence of instances of Activity carriedout in an organization with some specific objectives. If a process interacts with otherprocesses, it must participate in a Collaboration. The collaboration is a way of groupingseveral participants. Each Participant (aka Pool) must address only one process. Giventhe fact that a Participant is also an InteractionNode, it can send or receive several instancesof MessageFlow.
Figure 3.6 depicts the most instantiated meta-classes when is drawn a BPMN classdiagram. A FlowElementsContainer (which can be either a Process or a SubProcess) is acontainer of instances of FlowElement. A flow element can be either a FlowNode, a Se-quenceFlow or a DataObject. Instances of SequenceFlow can link various kinds of FlowNodeelements. A FlowNode can be one of the several different kinds of Activity, Event or Gate-way (see Figure 3.7).
55
3. ANALYSIS OF THE BPMN 3.3. BPMN Metamodel
Figure 3.4: BPMN abstract syntax – a subset of the BPMN metamodel
Figure 3.5: Process metaclass’ connections
56
3. ANALYSIS OF THE BPMN 3.3. BPMN Metamodel
Figure 3.6: Main meta-classes in a process orchestration
The ItemAwareElement is an abstract meta-class, from which derives several data re-lated meta-classes representing transient (DataObject) or persistent (DataStore) data con-tainers, as well as input or output data to/from Activity by means of meta-classes derivedfrom DataAssociation.
Figure 3.7: Derived meta-classes from FlowNode
3.3.2 Flaws in the Metamodel
The analysis of the BPMN metamodel, revealed a couple of flaws in the standard specifi-cation that are described below.
• The specification document allows the usage, by process modelers, of a visualshortcut that consists of the non-directional DataAssociation connected to a Sequence-Flow ( [BPM11] page 225) (see Figure 3.8 top). However, the metamodel only allowslinks among instances of the meta-classes DataAssociation and Activity (see Figure3.6). So, an instance of SequenceFlow cannot be directly linked to an instance ofDataObject via an instance of DataAssociation.
57
3. ANALYSIS OF THE BPMN 3.3. BPMN Metamodel
Tools that implement this visual shortcut, should instantiate the same meta-classesas if there was not the visual shortcut, i.e., implementing the regular solution thatconsists of drawing an instance of DataOutputAssociation going out from an instanceof Activity to an instance of DataObject. Conversely, an instance of DataInputAssoci-ation coming from the same preceding instance of DataObject goes to other instanceof Activity (see Figure 3.8 bottom).Furthermore, by permitting this kind of shortcut, the BPMN standard allows theoccurrence of ambiguous and misleading situations such as the one depicted onthe top of Figure 3.8, where DataObject1 is supposed to supply data to only oneof the instances of Task that follow Gateway1, as correctly depicted on Figure 3.8bottom;• The BPMN metamodel does not allow an instance of SubProcess to receive or send
instances of MessageFlow. This constraint introduces a huge limitation in the mod-ularization of BPMN processes’ through sub-processes, especially when there areinteractions among participants. Process modelers tend to ignore this constraint[Sil09], hence violating the metamodel as in Figure 3.9 where there are two instancesof MessageFlow connecting an instance of SubProcess to an external participant.For solving this issue in the BPMN metamodel, it would be enough that the is-arelationship between the metaclasses Task and InteractionNode (Figure 3.5), be re-placed by a is-a relationship between the metaclasses Activity and InteractionNode.However, the implementation of this rule at metamodel level, must be followedby another one implemented by tool makers in order to ensure consistency amongmodeling elements at different levels of detail. Therefore, it must be ensured thatthe instances of MessageFlow participating in interactions with the sub-process’s in-stance (Figure 3.9 top) level are the same that the one depicted interacting when thesub-process is detailed (see Figure 3.9 bottom).
Figure 3.8: A non-directional DataAssociation connected to a SequenceFlow
58
3. ANALYSIS OF THE BPMN 3.3. BPMN Metamodel
Figure 3.9: An instance of SubProcess receiving/sending instances of MessageFlow
3.3.3 Concepts not Covered & Proposed Extensions to BPMN
BPMN intends to model processes. However, BPMN constructs emphasize mainly thesupport of the control-flow and data perspective when expressing process orchestra-tions. As other process modeling languages, BPMN constructs have a shallow cover-age of resource or organizational aspects of process modeling. Moreover, BPMN focusesentirely on the behavioral aspect of the process model [Hol09] which is not consideredwide enough for defining and monitoring business objective and policy goal related tobusiness processes.
The BPMN execution semantics for process models intends to allow the definitionof executable processes, in order to give no room of interpretation on how to execute itby a process engine. An annex to the BPMN 2.0 specification has rules for transformingBPMN models into an executable BPEL format. However, since the internal structure ofBPEL is rather different from that of BPMN, such a transformation is still not easy andthere are also many problems in practice [Gao06, WDGW08]. Nevertheless, it is alreadypossible, through BPMN tools (e.g. Bizagi), to directly execute detailed BPMN modelsfor the purpose of simulation and monitoring [CT12].
Several initiatives have been triggered aiming to supplement known limitations ofBPMN and enhancing its capabilities. Some interesting proposals are: (1) Time-BPMN[GT09], an extension to BPMN that deals with the temporal perspective of processes;(2) BPMN-Q [Awa07] a language for querying the structure of processes, allowing toperform searches in processes, collecting information, finding specific elements, as wellas analyzing the behavior of a process; (3) xBPMN [Gro07], a revised formal control flowsemantics specified and validated against the requirements.
59
3. ANALYSIS OF THE BPMN 3.4. Weaknesses of the BPMN standard
3.4 Weaknesses of the BPMN standard
Aware of the BPMN standard’s limitations revealed in the assessment on the section 2.5.3,we wanted to confirm whether those limitations, were relevant enough to affect the qual-ity of process models produced using actual BPMN tools. This was done by checkingwhether current BPMN tools implementing the BPMN standard could rigorously verifyprocess models, and ultimately produce good quality artifacts.
If the lack of precise semantics of BPMN [BPM11] is in fact a major issue, this wouldnaturally be reflected in the outcomes of BPMN tools. The produced process modelswould reveal quality shortage, due to failures in BPMN tools interpreting and imple-menting BPMN well-formedness4 rules. Therefore the rules checking made by BPMN toolswould be unable of completely verify regular BPMN models.
In order to evaluate the extent in which currently available BPMN modeling toolsare supporting the generation of good quality process models, a survey was carried out.With this survey we intended to assess whether semantic rules regarding BPMN processmodeling, mostly specified in natural language in the standard documentation [BPM11],were effectively addressed and implemented by tool makers.
We tested the BPMN tools (represented by a sample of 11 current tools) to ensure aneffective verification of BPMN semantic rules (represented by a sample of 10 basic rules,either informally described in the BPMN standard documentation or by practitioners asbest-practices). The chosen rules are part of common process model diagrams, and mustbe used even by BPMN’s novices. So, it was naturally expected that all violations to therules could be captured by the sample of BPMN tools.
The set of chosen rules is summarized in section 3.4.1 and detailed in Appendix B.For testing the sample of BPMN tools, a model-snippet, depicted in Figure 3.10, was builtcontaining violations to the mentioned rules. The results regarding the effectiveness ofthe sample of current BPMN tools for verifying process models correctness are presentedin Table 3.1.
3.4.1 A Set of Rules for Assessing BPMN Tools
Among the 102 well-formedness rules we collected on more than five hundred pages ofthe BPMN standard document [BPM11], we chose a sample of 7 (7%) of them (see rules1-7 below) of the most used by every BPMN modeler in regular BPMN process modeling.We chose also a sample of 3 (10%) best-practices rules (see rules 8-10 below) advocatedby BPMN practitioners and experienced users [WM08, Sil09] from a total of 31. So, thenumber of the sample rules to be verified was a sample of 10 (8%) semantic rules, from atotal of 133.
4The term well-formedness is used here in the sense of static semantics [Aab96]. The static semanticsdefines restrictions on the structure of valid texts that are hard or impossible to express in standard syn-tactic formalisms, i.e., exclusively through the elements and relationships of the metamodel. For compiledlanguages, static semantics essentially includes those semantic rules that can be checked at compile time.
60
3. ANALYSIS OF THE BPMN 3.4. Weaknesses of the BPMN standard
1. A top-level Process can only be instantiated by a restricted set of Start Eventtypes (see Figure B.1).
Any container (Process or SubProcess) that does not have a parent container is con-sidered a top-level Process [BPM11, page 238] . Top-level processes can have oneof seven types of start events (see Figure 3.3, first column): none, message, timer,conditional, signal, multiple, and parallel [BPM11, page 112] .
2. Outgoing Sequence Flow not allowed in an End Event (see Figure B.2).
An end event indicates where a process will end. In terms of sequence flows, theend event ends the flow of the process, and thus, must not have any outgoing se-quence flow [BPM11, page 249].
3. Outgoing Message Flow not allowed in a Catch Event (see Figure B.3).
A start event or a catching intermediate event cannot have outgoing message flows[BPM11, page 251].
4. A Catch Event with incoming Message Flow must have Message or Multiple type(see Figure B.4).
A start event or a catching intermediate event with an incoming message flow mustbe of type Message or Multiple [BPM11, pages 44 and 271].
5. Explicit Start Event or End Event do not allow Activity or Gateway without in-coming/outgoing Sequence Flow (see Figure B.5).
Start Event and End Event are optional [BPM11, page 238]. However, if there is atleast one explicit start or end event in a container (Process or SubProcess), there mustnot be other flow nodes such as Activity and Gateway, without incoming/outgoingsequence flow [BPM11, pages 153, 289 and 430]. There are some exceptions: Com-pensation Activity and Event SubProcess do not have incoming and outgoing SequenceFlow.
6. A conditional Sequence Flow cannot be used if there is only one Sequence Flowout of the element (see Figure B.6).
If a conditional Sequence Flow is used from a source Activity, then there must be atleast one other outgoing Sequence Flow from that Activity. [BPM11, page 97].
7. A Boundary Event must have exactly one outgoing Sequence Flow, unless it hasthe Compensation type (see Figure B.7).
A Boundary Event is attached to an Activity and an outgoing exception flow comesout from it, through a Sequence Flow. Exactly one Sequence Flow is allowed from aBoundary Event except in the case that it is of type Compensation. In this particularcase an Association can replace or not the Sequence Flow [BPM11, pages 259, 440 and441].
61
3. ANALYSIS OF THE BPMN 3.4. Weaknesses of the BPMN standard
8. Use a Timer intermediate event with an Event Gateway (see Figure B.8).
One way for a modeler to ensure that a Process does not get stuck at an Event BasedGateway is to use a Catch Event of type Timer as one of the options to be reachedthrough a Sequence Flow leaving the gateway [WM08].
9. Use a Default Condition at an Exclusive Gateway (see Figure B.9).
One way for the modeler to ensure that the Process does not get stuck at an Exclu-sive Gateway is to use a default condition for one of the outgoing Sequence Flow.This creates a Default Sequence Flow. The latter is chosen if all the other SequenceFlow conditions turn out to be false [WM08].
10. Two activities in the same Process should not have the same name (see FigureB.10).
It is highly recommended that an activity name is unique. If it is required an activityto be reused in the process, a Global Activity should be used instead of duplicatingthe activity [Sil09].
For a visual perception and understanding of possible violations to the above rules,see Appendix B.
3.4.2 A Model-snippet for BPMN Tools’ Evaluation
Based upon the rules previously described (section 3.4.1), the business process model inFigure 3.10 was drew up inserting a violation of each one of the rules. A numbered labelwas put near the element(s) where the violation occurs.
The next step was to replicate the faulty business process model through the sample ofchosen eleven BPMN tools (see Appendix C), and collect data regarding caught violationsby each tool as summarized in next section (Table 3.1).
3.4.3 Results of Tools’ Assessment
Process modelers interest has been shifting from merely design tools to real processmodeling tools with verification capabilities, as well as repositories for models storage[HW11].
Since, in this dissertation, we were concerned mainly with the quality aspects of pro-cess modeling, we wanted to ascertain the effectiveness of current BPMN tools to ensurethe quality of the generated models. The set of tools from which the sample was chosenfor assessment, came from several origins, although mainly from the OMG site5. Froman initial list of 68 tools, a set of 11 units (16%) was chosen. In the selection process, weconsidered only process modeling tools. So we discarded graphic design tools, as well as
5http://www.bpmn.org/ [accessed in Feb. 10, 2013]
62
3. ANALYSIS OF THE BPMN 3.4. Weaknesses of the BPMN standard
Figure 3.10: Model-snippet for assessment of the effectiveness of BPMN tools verification
those with a main focus in processes’ execution at runtime (BPMS). We selected a conve-nience sample of BPMN tools, in order to have a mix of proprietary and open source tools,acknowledged by the market according the criteria of popularity and maturity.
Table 3.1: Rules’ violations detected on a model-snippet, by a sample of BPMN modelingtools
Adonis CE x x x x x x 5 71% 1 33% 60%Aris x 1 14% 0% 10%Bizagi x x 2 29% 0% 20%eClarus x 1 14% 0% 10%EA 0% 0% 0%iGrafx x x x x 4 57% 0% 40%MagicDraw x 1 14% 0% 10%Modelio x x x x x 4 57% 1 33% 50%Signavio x x x x x 5 71% 0% 50%TIBCO x x x x x x x 5 71% 2 66% 70%Visio+Modeler x x x x x 5 71% 0% 50%
Through the analysis of Table 3.1 and for the simple model-snippet checked, onecan conclude that none of the tools from the sample, fully detected the rules violations.The best score was attained by a unique tool that detected 70% (column Total %) of theviolations inserted in the model-snippet. Regarding the rules prescribed by the BPMNstandard the best score was given by tools that have detected 5 violations (71%) of the
63
3. ANALYSIS OF THE BPMN 3.5. Conclusion
standard’s rule violations. Concerning the violation of best-practices rules, the best scorewas given by a tool that detected 2 violations (66%) of the total of best-practices rules inthe model-snippet.
The achieved results seems to corroborate the idea that the limitations in the BPMNstandard, i.e. rules informally specified in natural language, contribute for poor qualityof process models, given the weakness of BPMN tools in terms of models’ verification.
3.5 Conclusion
We addressed the main characteristics of BPMN in section 3.2. The BPMN metamodelwas detailed (section 3.3.1) providing information regarding the main constructs used forbusiness process orchestration. We also highlighted flaws found in the BPMN metamodel(section 3.3.2), as well as proposed extensions to supplement BPMN (section 3.3.3).
A previous analysis of the BPMN (section 2.5.3) revealed some weaknesses amenableof harming the quality of produced process models. This evidence was corroboratedthrough a survey on BPMN tools (section 3.4) that confirmed limitations on using theBPMN to produce good quality models, due to the informal specification of rules in theBPMN standard. The relevance of the survey is given by the fact that the rules chosen tobe checked by the convenience sample of BPMN tools were some of the most basic onesthat a novice BPMN modeler would use.
64
4State-of-the-Art on Quality in Process
Modeling
"Do not go where the path may lead, go instead where there is no path and leave a trail."– Ralph Waldo Emerson
Context: Quality of process modeling can be addressed, when following a product-oriented approach, by assessing the quality of outcomes of process modeling activity,i.e., evaluating the internal and external qualities of process models.Objective: To survey existing approaches on assessing the quality of BPMN models,through the verification of correctness and measurement of quality characteristics.Method: A systematic review and a literature review were the main instruments used tocollect information and assess it using predefined dimensions.Results: A set of limitations of current research work were identified regarding BPMNmodels’ quality verification and measurement. Those issues were the base for formulat-ing the contributions of this dissertation.Limitations: Efforts were made to cover previous relevant research work related withthe domain of the present dissertation, by using namely a systematic review protocol.
65
4. STATE-OF-THE-ART ON QUALITY IN PROCESS MODELING 4.1. Introduction
Shallow coverage of some topics is supplemented by literature references.Conclusion: This chapter paved the way for the main contributions of the dissertation,by highlighting the main limitations of current approaches in addressing quality aspectsrelated with BPMN models.
4.1 Introduction
To support our own contributions, we will now survey previous research made regard-ing the quality of BPMN process models. As referred in section 2.4, we follow a product-oriented approach [MCN92] regarding the analysis of quality on BPMN models. There-fore, we focused on two quality perspectives: (1) the verification of BPMN models correct-ness; and (2) the measurement of quality characteristics of BPMN models.
Regarding the first perspective, quality verification, we were concerned in surveyingthe methods and tools used for formal verification of process models in general andBPMN models in particular. The aim was to review the specific approach used by thosemethods to check the quality characteristic of BPMN process models’ correctness.
In order to address the second perspective, the measurement of quality characteristics, itwas taken into account that, as mentioned in section 2.4.2, quality is a multi-dimensionalconcept, i.e., the quality model of a product or service should address different levels ofquality characteristics. So, regarding process models, the quality model should documentthe inter-relationships between quality characteristics and their measures1, as well as thelinks between internal and external quality attributes of process models. Therefore, theresults of directly measuring process models attributes (e.g. size, complexity) should belinked to the perceived quality of those models, such as correctness. For addressing thisquality perspective, the literature was surveyed in search of existent approaches dealingwith measurement of quality characteristics of process models in general and BPMN inparticular.
The structure of this chapter is the following: we surveyed in section 4.2 the generalapproaches to quality in process modeling from the verification (section 4.2.1) and mea-surement (section 4.2.2) perspectives. The approaches regarding quality of BPMN mod-els were specifically addressed in sections 4.3 and 4.4. It was carried out a systematicreview (section 4.3) to find and select studies concerning BPMN models’ quality. Thosestudies were classified according to a proposed taxonomy (section 4.3.2). Were also sur-veyed formal verification methods for checking BPMN process models’ properties andsemantics (section 4.3.3), as well as approaches for measurement of BPMN models qual-ity characteristics (section 4.4). Section 4.5 presents the main conclusions extracted fromthe state-of-the-art study.
1As said before the term measure is used instead of the term metric because more recent standards (e.g.ISO/IEC 25000 and ISO/IEC 15939 series) disregarded the use of the term metric.
66
4. STATE-OF-THE-ART ON QUALITY IN PROCESS MODELING 4.2. Quality on Process Modeling
4.2 Quality on Process Modeling
There is no general standard established for evaluating conceptual modeling [Moo05],and particularly the quality of process modeling and its outcome: process models (set ofdiagrams). However, due to the several process modeling languages available, there aresome dispersed research works about modeling guidelines [BRvU00, MRC07, VRM+08,MRvdA10] as well as the measurement of process diagrams characteristics [CMNR06,Car07, VCM+07]. Practitioners have also given some contributions, namely by promot-ing process diagrams’ modeling best-practices [WM08, Sil09, All10].
Meanwhile, a range of quality frameworks for conceptual modeling have been pro-posed in the literature, although none of them has reached a wide acceptance, thus be-coming a de facto standard. For instance, the SEQUAL framework [LSS94] provides asound theoretical basis for understanding quality in the conceptual modeling. SEQUALtakes the semiotic theory [Mor71] point of view, and has five components: the model, lan-guage, domain, audience participation, and perceived knowledge. Model quality is defined byrelationships between the model and the other four framework’s components in terms ofthe following models’ qualities: syntactic (model conformance to the language), semantic(model conformance to the domain) and pragmatic (model conformance to the audienceinterpretation) [Rec07]. The quality framework was empirically validated regarding pro-cess modeling [MSBS03]. The collected results raised questions about reliability of theframework to be applied in practice in its current form. Based upon the initial approachwas proposed an enhancement regarding process modeling [KSJ06]. Although the frame-work addresses quality in a systematic and comprehensive way, the drawback pointedout is too abstract to be used by practitioners [SD97].
4.2.1 Quality Verification
Model checking concerning process diagrams correctness has been a matter of intenseresearch. Some of the work has been done on the verification using modeling languageswith formal semantics (e.g. P/N). Due to the mathematical ground of those languages,they allow several formal verification methods, such as the verification of different classesof workflow definitions [vDA00]. However, for process modeling languages, which donot have a formal semantics and allow only processes’ informal representation, was re-quired a different approach to verification. Since process diagrams have to be translatedinto a specification to be executed by a machine, a general consensus was that process di-agrams had also to be formalized. Therefore, approaches for checking process diagramsfor semantic errors came to light, aiming process diagrams mapping to languages withformal semantics, such as the checking of EPC diagrams, using transformations to P/N,proposed in [LSW98, vdA99, DVDA04, vDvdAV05, Men07]. A modeling tool is providedby [KKGL10] and it applies graph-based rules for identifying problems in EPC processdiagrams.
67
4. STATE-OF-THE-ART ON QUALITY IN PROCESS MODELING 4.2. Quality on Process Modeling
The main characteristics of these approaches, excepting [KKGL10], is that the verifi-cation methods rules are only applicable after process diagrams are at a valid state, i.e.in a syntactically valid state, to be mapped from the specific process modeling languageinto the language and environment of the model checker.
4.2.2 Quality Measurement
There is no universally accepted measure for process diagrams quality but rather shouldbe considered a combination of measures [CMNR06]. This assumption has already beenreached by the software engineering community [NBZ06] regarding software productqualities [ISO11]. Therefore, efforts have been made in the process modeling field byproposing measures, as well as making empirical studies to call upon a set of measuresthat could measure and evaluate process diagrams characteristics, to provide informationto improve their quality attributes (section 6.2). However, it has been recognized that theresearch about process modeling, and particularly process diagrams quality attributes, isyet scarce [VCM+07].
Some of the most commons measures found for process diagrams were adopted fromthe software engineering area, given the closest resemblance between processes and com-puter programs namely in terms of the usage of algebraic techniques such as graph the-ory for analysis and representation [LK01]. In some sense, a process can be seen as acoarse grained view of a program, which is underpinned by specific constructs of pro-cess modeling (e.g. control-flow/selection, task/procedure) [CMNR06, GL07].
Recent surveys (e.g. [MGSA10]) summarized the most well-known proposals for pro-cess diagrams measures. These include the works of [LK01] (e.g. coefficient of networkcomplexity, complexity index, and restrictiveness estimator), [CMNR06] (number of ac-tivities, control-flow complexity, Halstead-based process complexity, and interface com-plexity), [MNvdA07] (size, separability, sequentiality, structuredness, cyclicity, and paral-lelism), [GL06] (number of activities, maximum/mean nesting depth, cognitive weights,information flow, knot-count, and anti-pattern counting) and [LvdA09] (extended Cardosometric, extended cyclomatic metric, and structuredness metric).
Some work has also been made in recent years, collecting empirical evidence thatcould support hypotheses regarding process diagrams measures. Particular interest hasthe work of Mendling, who focused in the EPCs process modeling language [MNvdA07].His aim was to establish a correlation among errors found in process diagrams and a setof proposed measures that disclose structural and behavioral characteristics of processdiagrams. Predictors were built to forecast errors in process diagrams based on mea-sures’ values. The results were supported by samples collected from process diagramsrepositories.
We consider as main characteristics of the aforementioned research done so far aboutprocess modeling measurement:
1. There is no standard definition of process modeling measures, resulting in different
68
4. STATE-OF-THE-ART ON QUALITY IN PROCESS MODELING 4.3. Verification of BPMN Models
algorithms for their computation;2. Some of the measures are defined in abstract terms, so the translation to a particular
process modeling language syntax can be done differently by distinct implemen-tors;
3. Measures are mostly used to support a posteriori analysis, i.e. historical analysisof process diagrams, rather than an analysis that could give real-time feedback,over the process diagram design phase. Hence, the measures are post-modelingmeasures instead of providing guidelines and preventing quality non-conformanceduring the modeling process;
4. There are no thresholds measures generally accepted, that could guide modelersalong process diagrams design, helping them to attain desirable levels of processdiagrams characteristics (e.g. complexity, modularity, size);
5. Distinct languages are used for process modeling and measures collection.
4.3 Verification of BPMN Models
In this and next section we will concentrate in the related work on quality of processmodeling in BPMN, which is the context of this dissertation.
In this section we will conduct a systematic review [Kit04] on BPMN process modelsverification. This systematic review aims to collect information regarding the approachesused for verification of BPMN models.
4.3.1 Systematic Review
The systematic literature review, as a scientific methodology, has been used on severalscientific domains (e.g. Software Engineering [BMNT05]) for integrating empirical re-search. By using the systematic review we want to identify, evaluate and interpret therelevant research upon BPMN models verification.
We started the systematic review by adopting a predefined review protocol to avoidthe possibility of bias (selection of individual studies not driven by our own expecta-tions). Therefore, we followed the methods specified in the protocol, including the iden-tification of the research question, the selection of studies, the quality assessment study,data extraction and monitoring, together with data synthesis.
The protocol underpinning our systematic review, is organized according to the fol-lowing activities:
1. Define a research question – Since this is qualitative research design we use inductivereasoning to propose the research statement (see section 4.3.2.1).
2. Locate and select relevant research studies – Without yet attempting an evaluation, wetried to find papers and reports in journals and papers with peer review. In section4.3.2.2 is detailed the research forums where was made the search, as well as theused search criteria.
69
4. STATE-OF-THE-ART ON QUALITY IN PROCESS MODELING 4.3. Verification of BPMN Models
3. Critically evaluate the studies– We assessed each research work against a set of criteriaregarding quality of BPMN process models (see section 4.3.2.4).
4. Combine the results –The findings were compiled and aggregate in Table 4.1 for com-parison purposes.
5. Publish the results – Besides being published in this section of the dissertation, theresults are intended to be published, in a near future, in a research paper for peervalidation purposes.
4.3.2 Protocol Instantiation
4.3.2.1 Research Question
The goal of the systematic review is to identify studies that provide an approach for thewell-formedness (semantic) rules verification of BPMN process models. This drives tothe research question of the current systematic review:
How BPMN process models are verified, regarding their quality?
4.3.2.2 Studies Selection
This step was intended to specify the search strategy aiming to detect as much as possiblerelevant literature on BPMN models’ verification. We were concerned in documentingthe whole search strategy so that other researchers can replicate the same review withidentical results.
To comply with the protocol, we had to search for documents using a predefinedquery string. We were interested in research works regarding exclusively to BPMN. Be-cause the year of BPMN’s first version was 2004 [BPM04], we have restricted the searchperiod from 2004 until the present moment. We also have chosen a set of terms relatedwith the compliance with the well-formedness of process models. The query string setup for querying the repositories was adapted to the specific syntax of each search engine,based on the following terms and boolean operators:
BPMN AND (verification OR checking OR formal OR semantic OR errors OR properties)
After the definition of the query string, the next step was to submit the query to dif-ferent sources of information. Those sources were: Google Scholar2, Microsoft AcademicResearch3, ACM Digital Library4, IEEE Xplore5, and Science Direct Digital Library6.
4. STATE-OF-THE-ART ON QUALITY IN PROCESS MODELING 4.3. Verification of BPMN Models
From each source we collected the 100 first items returned by the engine, ordered byits own relevance criteria. The 500 entries from all engines were stored in a databasebuilt for this purpose using EndNote7. Next, the database was queried to find duplicaterecords (articles found in more than one source). The result of this search was electedas the first set of articles. The set was composed of 55 studies found in more than onesource. Our assumption for the first cut-off of research works to be reviewed was definedas follows: if a paper was referenced by more than one search engine, it meant that it wasmore relevant than the others referenced by only one search engine.
After selecting the first set of studies, we performed a new selection by assessing thestudies’ content. The idea was tuning the set of initial research works by applying a sec-ond selection criteria: containing information about any kind of approach concerning theverification of process models, i.e., the article should refer the detection of non confor-mances due to semantic errors or properties’ faults (e.g. deadlock). This second selectionwas done by the reading of the paper and the search, for the verification approach, in theabstract, introduction and conclusions of the paper.
As a result of applying a more narrow criteria in the second selection, the final set wascomposed only by 13 articles (Table 4.1).
4.3.2.3 Synopsis of Selected Research Works
To discover the proposal made in each of the research works, we analyzed how was car-ried out the verification of BPMN models. We summarize below each of the approachesmentioned in Table 4.1:
1. The article "A pattern-based approach for the verification of business process de-scriptions" [PS13] presents an approach for verifying business process descriptionspresented in any style (e.g. as text, tables or graphical artifacts), created with pro-cess modeling languages such as BPMN. Because of its generality and the avail-ability of tools, model checking is used to verify business process descriptions,particularly the SPIN model checker. The proposed composition-based approachpermits the semi-automatic implementation of business process description in theSPIN tool and the verification of numerous correctness properties, which refer toworkflow control flow patterns, safety and liveness. For the basic patterns andfragments, PROMELA in-line constructs are provided, and is suggested the set ofapplicable correctness properties. The correctness properties are specified as tem-plates in linear temporal logic. Implementing a business process description con-sists of assembling the in-line constructs and associating business semantics withthe symbols in the logical formulas of the correctness properties. For verificationare used the SPIN algorithms. By using the presented approach, business processdescriptions can be checked for correctness.
7http://endnote.com/
71
4. STATE-OF-THE-ART ON QUALITY IN PROCESS MODELING 4.3. Verification of BPMN Models
Table 4.1: Research Works Selected# Title Authors
1 A pattern-based approach for the verification ofbusiness process descriptions [PS13]
S. Patig, M. Stolz
2 A visual token-based formalization of BPMN2.0 based on in-place transformations [GD13]
P. Gorp, R. Dijkman
3 Adding Preciseness to BPMN Models[CBeA12]
A. Correia, F. Abreu
4 An eclipse plug-in for formal verification ofBPMN processes [FABD10]
C. Flavio, et al.
5 Analysis on demand: Instantaneous soundnesschecking of industrial business process models[FFK+11]
D. Fahland, et al.
6 Developer-friendly verification of process-based systems [PFS10]
E. Pulvermueller, et al.
7 Direct verification of BPMN processes throughan optimized unfolding technique [FPPR12]
D. Falcioni, et al.
8 Formal analysis of BPMN models using Event-B [BW10]
J. Bryans, W. Wei
9 Formalisations and Applications of BPMN[WG11a]
J. Gibbons, P. Wong
10 On the refactoring of activity labels in businessprocess models [LSM12]
H. Leopold, et al.
11 Property specifications for workflow modelling[WG11b]
P. Wong, J. Gibbons
12 Semantics and analysis of business processmodels in BPMN [DDO08]
R. Dijkman, et al.
13 Visually specifying compliance rules and ex-plaining their violations for business processes[AWW11]
A. Awad, et al.
2. The paper "A visual token-based formalization of BPMN 2.0 based on in-placetransformations" [GD13] provides a BPMN semantics formalization that consistsof in-place graph transformation rules that are documented visually using BPMNsyntax. In-place transformations update models directly and do not require map-pings to other languages. A tool and test-suite was used to develop a referenceimplementation of all rules. The formalization intends to complement the stan-dard, in particular because the rules have been extensively verified and because isfacilitated the conceptual validation (the informal semantics also describes in-placeupdates).
3. In our article "Adding Preciseness to BPMN Models" [CBeA12], OCL invariants[OMG06] were formalized and added to OMG’s BPMN metamodel specification.Those invariants were related to well-formedness rules informally described inBPMN specification, as well as rules regarding models’ properties. In this paper it
72
4. STATE-OF-THE-ART ON QUALITY IN PROCESS MODELING 4.3. Verification of BPMN Models
was briefly described how the approach was operationalized, namely by develop-ing several transformations to allow checking rules conformance of process modelsproduced with a BPMN modeling tool. A metamodel-based checking facility wasalso developed as a JUnit test-suite. Each test case checked the validity of a modelsnippet implementing a specific BPMN rule. Using the same metamodel-based ap-proach, it was also formalized a set of best practices for BPMN modelers, based onpublished recommendations produced by BPMN practitioners.
4. The article "An eclipse plug-in for formal verification of BPMN processes" [FABD10]proposes a novel approach enriching the design with a formal semantic and a sys-tematic integration of formal verification. The approach has been implemented as atool that allows verifying business processes. The tool (BP4PA)is an Eclise plug-inthat uses BPMN notation for business process specification, implements a map-ping from the BPMN to the CSP formal language and supports CSP verification viamodel checking.
5. In "Analysis on demand: Instantaneous soundness checking of industrial businessprocess models" [FFK+11] models are investigated for soundness (absence of dead-lock and lack of synchronization) using three different approaches: the businessprocess verification tool Woflan, the Petri net model checker LoLA, and a techniquebased on SESE decomposition. The various techniques used by these approachesare evaluated in terms of their ability of accelerating the check. Soundness is usedfor example as a precondition to map a process modeled in a graph-based languagesuch as BPMN, into BPEL in a way that preserves the execution semantics and thestructure of the process.
6. In the article "Developer-friendly verification of process based systems" [PFS10] achecking system is presented that integrates a graphical notation for a user-friendlyspecification and an extended specification language together with a correspondingverifier which supports the checking of many different types of elements. The inte-gration is realized by an XML-based transformation system which links the graph-ical editor of a process modeling such as BPMN, P/N or EPC, to the checking tool.
7. The work "Direct verification of BPMN processes through an optimized unfoldingtechnique" [FPPR12] is a Java based verification approach for BPMN. In particular,a precise Java mapping is defined for the main elements of the BPMN notation.The relations among the different elements of a BPMN specification are supportedby the inclusion of specific attributes and methods in the created Java objects. Thebehavior of a set of BPMN interrelated objects can be explored using an algorithmdefined for that purpose. Such an algorithm permits to avoid the state explosionphenomenon using an ad-hoc unfolding technique. It has been developed a plug-infor the Eclipse IDE platform. It permits to have an integrated environment in whichis designed a business process, to verify it, and to check the result of the verificationin order to improve the business process itself. This iterative approach can continueuntil all the issues highlighted by the verifier are solved.
73
4. STATE-OF-THE-ART ON QUALITY IN PROCESS MODELING 4.3. Verification of BPMN Models
8. "Formal analysis of BPMN models using Event-B" [BW10] demonstrates how for-mal verification may add value to the specification, design and development ofbusiness process models in an industrial setting. The analysis of models is achievedvia an algorithmic translation from the BPMN to Event-B, a formal language sup-ported by the Rodin platform which offers a range of simulation and verificationtechnologies.
9. The article "Formalisations and Applications of BPMN" [WG11a] provides a frame-work for precise specifications and formal verifications of workflow processes mod-eled as BPMN diagrams. Two behavioral semantics are provided for BPMN in theprocess algebra CSP. Existing CSP refinement orderings are applied to both the re-finement of business process diagrams and the verification of behavioral compat-ibility of business process collaborations. The first semantic model is an untimedmodel, focused on the control flow and communication of business processes. Thesecond semantic model extends the first one to capture the timing aspect of behav-ior. It is also considered the applications of the semantic models. The secondaryobjective of the approach was to apply BPMN and the semantic models to reasonabout long running empirical studies (e.g. laboratory experiments, clinical trials).
10. "On the refactoring of activity labels in business process models" [LSM12] addressthe problem of activity label quality in business process models. The technique wasdesigned for the recognition of labeling styles, and the automatic refactoring oflabels with quality issues. It was developed a parsing algorithm that is able to dealwith the shortness of activity labels, which integrates natural language tools likeWordNet and the Stanford Parser. Using three business process model collections,one of them in BPMN, from practice with differing labeling style distributions, itwas demonstrated the technique’s applicability. The technique shifts the boundaryof process model quality issues that can be checked automatically from syntactic tosemantic aspects.
11. The paper "Property specifications for workflow modelling" [WG11b] considersa pattern-based approach to expressing behavioral properties, supplementing thedifficulties of BPMN to construct behavioral properties against to which modelsmay be verified. A property specification language is described (PL) for capturinga generalization Property Specification Patterns, and presented a translation fromPL into a bounded, positive fragment of linear temporal logic, which can then be au-tomatically translated into CSP for simple refinement checking. These results leadto a compositional approach for ensuring deadlock freedom of interacting businessprocesses. This work follows a previous one of the authors, providing formal be-havioral semantics for the BPMN in the process algebra CSP.
12. "Semantics and analysis of business process models in BPMN" article [DDO08] pro-poses a mapping from BPMN to a formal language (Petri nets) for which efficientanalysis techniques are available. The proposed mapping has been implementedas a tool that, in conjunction with existing Petri net-based tools, enables the static
74
4. STATE-OF-THE-ART ON QUALITY IN PROCESS MODELING 4.3. Verification of BPMN Models
analysis of BPMN models.13. The article "Visually specifying compliance rules and explaining their violations
for business processes" [AWW11] utilizes a visual language, BPMN-Q, to expresscompliance requirements visually in a way similar to that used by business expertswhen building process models. Still, using a pattern based approach, each BPMN-Q graph has a formal temporal logic expression in Computational Tree Logic (CTL).Moreover, the user is able to express constraints, i.e., compliance rules, regardingcontrol flow and data flow aspects. In order to provide valuable feedback to a userin case of violations, the approach depend on temporal logic querying approachesas well as BPMN-Q to visually highlight paths in a process model whose executioncauses violations.
4.3.2.4 Analysis and Findings
After the set of selected research works that have been revised, this section presents thefindings. The studies were classified according to a number of dimensions, that we deemrelevant for being satisfied by a comprehensive approach, claiming to be able to checkBPMN process models. Those dimensions are:• Well-formedness (Semantic) Checking – is concerned whether the approach encom-
passes the checking of well-formedness rules prescribed on the BPMN standarddocument [BPM11] (Appendix E);• Properties Verification – is related to the suitability of the approach for checking
domain-independent properties (e.g. deadlock, liveness);• Measurement – whether the approach provides measures regarding to the internal
and external attributes of the models, contributing for the improvement of processmodels’ quality;• Modeling Guidance – is related to the capability of the approach to provide direct
guidance to the process modeler, in design time, even in the presence of incompleteprocess models;• Empirical Validation – is related to the effectiveness of the approach evidenced through
statistically grounded empirical studies.A score was assigned to each study in each dimension, based on an ordinal scale
depicted as a symbol: None/Basic – O ; Partial/Moderate – ; Full/Plenty – .Table 4.2 provides the synthesis of the systematic review results.
75
4. STATE-OF-THE-ART ON QUALITY IN PROCESS MODELING 4.3. Verification of BPMN Models
Table 4.2: Classification of the selected studies
# Research WorkWell-formedness
CheckingProperties
VerificationMeasu-rement
ModelingGuidance
EmpiricalValidation
1 A pattern-based ... O O O O2 A visual token-b... O O O 3 Adding Precisene... O O4 An eclipse plug-... O O O 5 Analysis on dema... O O O O6 Developer-friend... O O O O7 Direct verificat... O O O8 Formal analysis ... O O O O9 Formalisations a... O O O
10 On the refactori... O O O O11 Property specifi... O O O O12 Semantics and an... O O O O13 Visually specify... O O O O
As can be seen in the previous table, most of the works address essentially the prop-erties verification dimension. Our approach (the third in Table 4.2) published in[CBeA12] seems to be the one that encompasses more dimensions. In this disser-tation we will delve into the "Semantic Checking" and "Modeling Guidance" dimen-sions (chapter 5). We will also tackle the "Measurement" dimension in chapter 6, aswell as "Empirical Validation" dimension in chapters 8 and 9. By doing that we willbe able to cover all the four dimensions in detail. The full coverage of the dimension"Properties Verification" is left for tools and techniques already existent, since theyare ensuring already a good coverage of it.
4.3.3 BPMN Formal Verification Methods
In the previous systematic review we overviewed some of the relevant works surveyedregarding verification on BPMN process models. Next we summarize some of the mostwell-known formal verification methods that are used on checking of BPMN models,through the specific formalisms detailed in Appendix A (section A.2).
• Communicating Sequential Processes (section A.2.1) – Wong and Gibbons [WG08, WG11a]use the CSP language and the behavioral semantics as the denotational model foraugmenting the semantic model of BPMN, introducing relative timing information,and then allowing the specification of timing constraints on concurrent activities.The processes’ behavioral properties of BPMN diagrams are analyzed using theFDR model checker.• Petri Nets (section A.2.2) – Dijkman et al. [DDO07] provide a formal semantics for a
subset of BPMN constructs in Petri nets. In particular, they provide a semantics thatmaps BPMN pools into Workflow nets. A Workflow net is a P/N such that there is
76
4. STATE-OF-THE-ART ON QUALITY IN PROCESS MODELING 4.3. Verification of BPMN Models
a unique source place i (•i = ∅), a unique sink place o (o• = ∅), and every other placeand transition is on a directed path from the unique source place to the unique sinkplace.The semantics only encompasses a subset of BPMN, and does not properly modelmultiple instances, exception handling and message flows.• Web Ontology Language (section A.2.3) – A common property of ontologies and the
OWL semantics is the so-called open-world assumption [HPSVH03], a form of par-tial description or under-specification as a means of abstraction, i.e., from the ab-sence of statements, a deductive reasoner must not infer that the statement is false.In [Nat11] it is defined an ontology that formally represents the BPMN specification(BPMN 2.0 Ontology), and can be used as a knowledge base. The description of anelement is combined within the corresponding class and further explanations areprovided in annotations. This is claimed to allow a much faster understanding ofBPMN. In addition, the ontology is used as a syntax checker to validate concreteBPMN process models.• Abstract State Machines (section A.2.4) – The approach regarding BPMN semantics
in [BS11], uses ASM [BT08], as a form of rigorous pseudo-code that follows theinheritance steps in the process modeling language class hierarchy. The abstractmodel of the dynamic semantics8 of the BPMN language is attained, by insertingrules as behavioral elements at appropriate places in the class hierarchy, definingtherefore, the language’s execution semantics. The enhanced process modeling lan-guage model can be used to check the conformance of process diagrams.
4.3.4 Conclusions on Verification of BPMN Models
We highlight the following main limitations of above proposals regarding BPMN modelverification:
1. Most of the verification methods rules are only applicable, after process diagramsare at valid state to be mapped from a specific process modeling language into thelanguage and environment of the model checker.
2. The approaches presented, are technically demanding w.r.t. the formalisms, evenfor advanced process modelers, from whom one can hardly expected to be ac-quainted with the above mentioned formalisms. Moreover, it must be made avail-able a blended environment integrating a process modeling language tool and amodel checker tool.
3. The ontological approach proposed uses the open-world assumption [HPSVH03], adifferent assumption used by the MDA approach (see chapter 7), which advocatesthat what is not currently known to be true, is false, and therefore assumes that the
8The dynamic semantics (aka execution semantics) of a language defines how and when the variousconstructs of a language should produce a program behavior.
77
4. STATE-OF-THE-ART ON QUALITY IN PROCESS MODELING 4.4. Measurement of BPMN Models
model has complete information to restrict arbitrary extensions of the system thatcould lead to inconsistencies (closed-world assumption).
The proposed approach to get over the mentioned limitations is to supplement thestandard process modeling language with rules that could enhance BPMN models’construction. Being an integral part of the BPMN metamodel, BPMN semantic rulesimplemented as OCL invariants, could be enforced by the same tools that alreadysupport the BPMN modeling language. Therefore, during the BPMN process modeldesign, in addition to syntax error verification, which is available to a certain degreefrom current tools, the modelers would have also, real-time checking about seman-tic violations in models (e.g. a throw event without a corresponding catch event; amismatch between flows from a split parallel gateway and the incomings on the cor-responding joint element, a possible cause of a deadlock situation). Similarly, BPMNmodeling best-practices of an organization, could also be established and enforced.This kind of rules (standard or from best-practices) could greatly enhance the qualityof BPMN diagrams resulting from the design phase of process’s life cycle.
4.4 Measurement of BPMN Models
In this section we survey the literature in search for contributions on proposed measuresfor evaluate internal and external qualities of process diagrams using BPMN.
Few research works have been published addressing the measurement of BPMNmodels’ quality characteristics. Rolón et al. performed an empirical study in which ob-servations of a complexity measure were correlated with process diagrams qualities ofunderstandability and modifiability [RCG+09].
In other approaches9, BPMN elements’ simple counting measures were defined andimplemented using OCL [RRG+09]; and several measures, proposed in the literature forprocess diagrams, were also implemented through OCL [BeAPFC10]. Both approacheswere formalized using the MetaModel-Driven Measurement (M2DM) approach [BeA01].
The two last works mentioned have some resemblance with the approach proposedin this dissertation. However the mentioned proposals, do not theoretically validate (seechapter 6) the previously proposed measures. Moreover, the measures were computedwithout considering the syntactical and well-formedness verification of BPMN diagrams.
9These approaches were based upon ad-hoc metamodels of BPMN v1 , since there were not an officialmetamodel version released.
78
4. STATE-OF-THE-ART ON QUALITY IN PROCESS MODELING 4.5. Conclusion
To overcome the above mentioned limitations, we propose a set of BPMN measures,upon the current BPMN v2 standard. So, the idea is to extend the OMG’s BPMNmetamodel with rigorously defined measures, in a declarative way using OCL. Themeasures as an integral part of the BPMN, will be implemented by tools supportingthe BPMN modeling language. Therefore, modelers will have real-time measuresabout process diagrams characteristics. These measures could be compared withgeneral or organizational thresholds, and will serve as guidance for attaining qual-itatively better process diagrams. Furthermore, it will facilitate the data collectionwithin the modeling process for statistical analysis and quality assurance.
4.5 Conclusion
This chapter intends to summarize the research work done in the domain in which thedissertation is expected to give some contributions, which are presented in chapters 5and 6 and validated in chapters 8 and 9.
The chapter addresses, following a product-oriented approach, the quality of processmodels in two perspectives: (1) concerned with the verification of process models’ cor-rectness; and (2) with the measurement of quality characteristics of process models. Bothperspectives of quality are taken for process models in general (section 4.2) and BPMNmodels in particular (sections 4.3 and 4.4).
It was conducted a systematic review regarding BPMN process models verification. Itwas concluded that most of the research works surveyed addressed essentially the Proper-ties Verification aspects, i.e., the papers were related with checking of domain-independentproperties (e.g. deadlock, liveness). Other aspects considered relevant for this disserta-tion were insufficiently covered, namely the Well-formedness Checking and Modeling Guid-ance aspects (to be considered in this dissertation in chapter 5), as well as the Measurementdimension (to be covered in chapter 6), and Empirical Validation dimension (chapters 8and 9).
It was also done a literature review, to summarize some of the most well-known for-mal verification methods that have been used on checking BPMN models (CSP, P/N,OWL, ASM). The limitations, from our point of view, of the methods were highlightedand was made a proposal to overcome those limitations: supplementing the standardprocess modeling language with rules, implemented as OCL invariants, that could en-hance BPMN models’ verification in design time.
Lastly, it was made another literature review regarding the research work done aboutmeasurement of BPMN Models quality characteristics. The limitations pointed out to thesurveyed work are: (1) not validating theoretically, the proposed measures; as well as (2)the computation of measures without considering the syntactical and well-formednessverification of BPMN diagrams. To overcome these limitations, it was proposed the elic-itation of a set of BPMN measures, to be aligned with the current BPMN standard.
79
[This page is intentionally blank]
5Verification of BPMN Models
"The man of science has learned to believe in justification, not by faith, but byverification"
Context: The BPMN specification has rules regarding the correct usage of the language’sconstructs, which are informally defined yielding dubious interpretation. There are alsorules that came from disseminated best-practices both from academics and practitioners.Not infrequently process modelers violate BPMN underlying grammatical rules, usingconstructs outside of their designated scope.Objective: To help BPMN process modelers to tackle the complexity of the BPMN lan-guage and produce well-formed models, we propose to formalize both the standard andbest-practices rules.Method: The rules formalization is implemented with the OCL invariants upon theBPMN metamodel.Results: A set of verification rules enforcing the compliance of BPMN models with boththe BPMN specification and BPMN modeling best-practices.Limitations: Since properties verification (e.g. deadlock, liveness) is a topic alreadycovered by several other approaches, it is not targeted in the present work.
81
5. VERIFICATION OF BPMN MODELS 5.1. Introduction
Conclusion: The extended BPMN metamodel has embedded the well-formedness rules,besides the abstract syntax of the process modeling language, which allows the verifi-cation of BPMN models correctness.
5.1 Introduction
The popularity of BPMN stems primarily from the rich expressiveness of its graphicalrepresentation. Activities in a process and their execution constraints expressed graphi-cally eases the communication about processes among the different involved stakehold-ers. However, there are other aspects important for a process modeling language, notavailable in BPMN metamodel: the rigorous definition of rules that BPMN models mustconform with.
The BPMN metamodel [BPM11] gives an abstract syntax for the constructs of BPMN(section 3.3). This is described at the meta-level using a class diagram. The BPMN meta-model can serve as a precise description of the notation and is therefore useful in imple-menting modeling tools, since it can be used as a basis to define the language syntax.However, it cannot serve as a precise description of the meaning and usage of BPMNconstructs.
Despite the strengths of the BPMN standard, its use on process modeling projects wasnot reached yet its potential. BPMN well-formedness rules are given in plain text so theyare subjected to different interpretation by process modelers and tool makers. The lackof a rigorous language semantics for the underlying modeling notation is a significantsource of problems. A consequence of this, given the expressiveness of the language, isthat it is not difficult for process modelers to generate faulty models (see Chapter 8). Thelack of precise semantics for BPMN makes difficult to produce well-formed models. Ithas been also difficult to develop effective tools for verification procedures (as surveyedin section 3.4).
The level of formality of a model is not absolutely determined by its form of repre-sentation. Particularly, graphical notations can be seen as formal if they possess a pre-cise semantics associated to their constructs. In [FELR98] three general approaches areidentified to formalize modeling concepts. In this dissertation we follow the so calledsupplemental approach which uses formal statements to replace parts of an informal modelthat is expressed in natural language.
A formal semantics for the BPMN standard can be obtained by defining a mappingfrom syntactic structures in the informal modeling domain to artifacts in the formallydefined semantic domain [FELR98]. This mapping, often called a meaning function, isused to build interpretations of the informal models. Rather than generate formal spec-ifications from informal BPMN models and requiring that developers manipulate theseformal representations, our approach is to provide formal semantics for graphical mod-eling notation and then provide a tool that allow modelers to directly check the BPMNmodels they have created. Defining meaning functions allow exploring the appropriate
82
5. VERIFICATION OF BPMN MODELS 5.2. BPMN Rules Formalization
formal semantics for graphical modeling constructs. BPMN tools (and not the processmodelers) should use these mappings to verify the correctness of the BPMN models.
A central part of this dissertation is to develop a formal version of BPMN rules thatcould be used to build precise and analyzable models. Using the OCL textual notation,a BPMN model can be analyzed to determine the truth or falsity of some property beingasserted about the process model.
Each analysis involves applying a sequence of invariants to the process model to de-rive the required conclusion. After the specification of the invariants, and since BPMNmetamodel alone is not expressive enough to define all properties, we must implementthe invariants, to supplement the metamodel.
Previous formalization of modeling notations [EFLR98] indicate that a precise anduseful semantics must be complete (i.e., meanings must be associated with each well-formed syntactic structure), the intended level of abstraction must be preserved (i.e., theelements in the semantic domain must be at the same level of abstraction as their corre-sponding modeling concepts), and the final result must be understandable by develop-ers. Based on these guidelines, we formalized for BPMN a set of static semantic rules1
[Aab96]. With OCL we were able to improve the static semantics of BPMN within theUML metalanguage context.
Together with the BPMN graphical notation, to assist the development of processmodels, the static semantic (well-formedness) rules contribute to the preciseness of spec-ification and to assist developers in moving towards correct implementation of processmodels. The formal semantics also allows automated tool to better support the notation.
The informal usage of BPMN as a modeling technique, is amenable to produce mod-els hard to read and interpret, due to the difficulty to understand used BPMN constructsby process modelers. This problem highlights the utility of formal techniques for convey-ing rigorous interpretation of the constructs. In the end it will be expected that BPMNmodels could be amenable to a more rigorous checking and analysis, through a simulta-neous syntactical and well-formedness validation using a tools such as USE (UML basedSpecification Environment) [GBR07].
In section 5.2 is presented a formalization of BPMN well-formedness rules, as well asits implementation using OCL. These well-formedness and best-practices rules attachedto the BPMN metamodel will serve as constraints for the verification of BPMN models.Section 5.3 summarizes the chapter.
5.2 BPMN Rules Formalization
The BPMN specification [BPM11] contains the rules regarding the correct usage of thelanguage’s constructs, scattered by a document of five hundred of pages. Those rules
1 The static semantics defines restrictions on the structure of valid texts that are hard or impossible toexpress in standard syntactic formalisms, i.e., exclusively through the elements and relationships of themetamodel.
83
5. VERIFICATION OF BPMN MODELS 5.2. BPMN Rules Formalization
are expressed in natural language, yielding sometimes dubious interpretation. On theother hand, there are also rules that came from disseminated best-practices both fromacademics and practitioners [WM08, Sil09, All10]. In order to help BPMN process mod-elers to tackle the complexity of the process modeling language, we will formalize boththe well-formedness and best-practices rules. Those rules are to be included in the BPMNmetamodel to support the process diagram’s verification, enforcing well-formedness rulescoming either from the BPMN specification or from the current BPMN modeling best-practices.
For implementing the rules we will use the OCL [OMG06]. OCL is a declarativeand predicate logic like language that supplements the UML, and is used to implementthe rules by means of invariants. A formal language, such as the OCL, can contributefor rigorously expressing well-formedness rules, which are hard to convey by graphicalnotations. With OCL we are able to improve the static semantics of BPMN, framed by theBPMN metamodel (section 3.3). Supplementing BPMN metamodel, with formal ruleswill help modelers to attaining well-formed BPMN process models.
5.2.1 Well formedness Rules
In Appendix E is listed the set of well-formedness rules withdrawn from the processmodeling language specification [BPM11], through a rule mining process. They were scat-tered by the text and tables of the document standard in natural language.
To give an idea how the specification of well-formedness rules are buried in the stan-dard’s document we reproduce an excerpt, concerned with rule of section 5.2.1.1. Therule is in Table 10.88 - End Event Types on [BPM11, page 248], and is concerned with theEscalation Intermediate Event.
This type of End indicates that an Escalation should be triggered. Other activethreads are not affected by this and continue to be executed. The Escalation willbe caught by a Catch Escalation Intermediate Event with the same escalationCode orno escalationCode which is on the boundary of the nearest enclosing parent Activ-ity (hierarchically). The behavior of the Process is unspecified if no Activity in thehierarchy has such an Escalation Intermediate Event.
All the rules are implemented in OCL2. In sections 5.2.1.1 – 5.2.1.3 we present a sam-ple of 3 well-formedness rules. Besides the rationale of the rule, it is also presented adepiction of the correct and incorrect usage of the well-formedness rule, as well as itsimplementation as OCL invariant.
2See http://sdrv.ms/16EvDjG
84
5. VERIFICATION OF BPMN MODELS 5.2. BPMN Rules Formalization
5.2.1.1 A Throwing Escalation Intermediate Event matches a non-Interrupting Esca-lation Catch Event
If an intermediate escalation event does not abort its current activity, the attached bound-ary event of the activity must be either a non-interrupting event with the same nameof the intermediate escalation event, or unnamed. In this case, the throwing escalationevent in the sub-process needs to be an intermediate event. So, an Escalation intermedi-ate Event needs a non-interrupting Catch Escalation Event to capture it [All10, BPM11,page 248].
Figure 5.1: A Throwing Escalation Intermediate Event matches a non-Interrupting Esca-lation Catch Event
Correct: A Throw Escalation Intermediate Event Escalation2 (top) is caught by an non-Interrupting Escalation Event Escalation2 (middle).
Wrong: A Throw Escalation Intermediate Event Escalation2 is caught by an InterruptingEscalation Event Escalation2 (bottom).
A well-formedness rule can be enforced by attaching the following invariant to theThrowEvent element of the BPMN metamodel.
85
5. VERIFICATION OF BPMN MODELS 5.2. BPMN Rules Formalization
Listing 5.1: A Throwing Escalation Intermediate Event matches a non-Interrupting Esca-lation Catch Event.
5.2.1.2 A flow from an Interrupting Catch Event must merge the normal flow throughan Exclusive Gateway
An attached interrupting event simply creates a token if it occurs during the execution ofthe corresponding activity. If both sequence flows (normal and interrupting) should bemerged by a gateway, an exclusive gateway would be required.
When an activity is carried out, an exception path originated from the InterruptingCatch Event can be taken due to early abortion of the activity. If later the exception pathjoins the normal sequence flow of the Process, it requires an Exclusive Gateway [All10,BPM11, pages 258, 259].
A well-formedness rule can be enforced by attaching the following invariant to theBoundaryEvent element of the BPMN metamodel.
Listing 5.2: A flow from an Interrupting Catch Event must merge the normal flowthrough an Exclusive Gateway.
13 -> flatten())->select((oclIsTypeOf(ExclusiveGateway) or
14 oclIsKindOf(Activity) or oclIsTypeOf(EndEvent)) and
15 oclAsType(FlowNode).isJoin())->size()>0)
86
5. VERIFICATION OF BPMN MODELS 5.2. BPMN Rules Formalization
Figure 5.2: A flow from an Interrupting Catch Event must merge the normal flow throughan Exclusive Gateway
Correct: If the event TimeOut occurs, Activity1 will be aborted, and the downwards out-going exception flow will be taken. Later it will join the normal flow of the processvia an Exclusive Gateway (top).
Wrong: The exception flow joins the normal flow through an Inclusive Gateway (bot-tom).
5.2.1.3 An Event SubProcess must not have any incoming or outgoing Sequence Flow
An Event SubProcess must not have incoming or outgoing Sequence Flow, since it is theStart Event that triggers the sub-process execution [BPM11, page 176].
The well-formedness rule regarding incoming or outgoing sequence flows can be en-forced by attaching the following invariant to the SubProcess element of the BPMN meta-model.
Listing 5.3: An Event SubProcess must not have any incoming or outgoingSequence Flows.
1 context SubProcess
2 inv noIncomingAndOutgoingSequenceFlow:
3 self.isEventSubProcess()
4 implies
5 (self.inputSequenceFlows()->isEmpty() and
6 self.outputSequenceFlows()->isEmpty())
87
5. VERIFICATION OF BPMN MODELS 5.2. BPMN Rules Formalization
Figure 5.3: An Event SubProcess must not have any incoming or outgoing SequenceFlows
Correct: An event sub-process without incoming or outgoing Sequence Flows (top).
Wrong: An event sub-process must not have any incoming or outgoing Sequence Flows(bottom).
5.2.2 Best-practices Rules
In this section we included some rules that we mined from works of academics andpractitioners3 [WM08, Sil09, All10], aiming to ensure that common patterns and goodpractices are followed when process models are built using BPMN. These rules are notmandatory and are heavily dependent upon process modeling and methodological pro-cedures implemented in each organization.
All the rules are implemented in OCL4. In the sections 5.2.2.1 – 5.2.2.3 we present asample of 3 best-practices rules. In order to distinguish the invariants implementing best-practices rules, from the ones prescribed by BPMN well-formedness rules, we prefixedthe former invariants’ names with ’bp ’. The following sections present the rationale ofthe rule, a depiction of the correct and incorrect usage of the rule, as well as its imple-mentation as OCL invariant.
5.2.2.1 Use explicitly Start Events and End Events
Process modeling best practices recommendations advise the explicit use of start and endevents [WM08].
3There are some sites with lists of rules claimed by practitioners as essential to be followedin the BPMN process modeling. One of the most known examples is the Bruce Siver site athttp://www.brsilver.com/2010/09/28/the-rules-of-bpmn/ [accessed in April, 16th 2012]
4See http://sdrv.ms/16EvDjG
88
5. VERIFICATION OF BPMN MODELS 5.2. BPMN Rules Formalization
Figure 5.4: Use explicitly Start Events and End Events
Use: Explicit usage of Start and End Event (top).
Avoid: Use of instances of FlowNode as implicit Start and End Event (bottom).
The best-practice rule regarding Start and End Event can be enforced by attachingthe following invariant to the FlowElementsContainer element of the BPMN meta-model.
Listing 5.4: Use explicitly Start Events and End Events.1 context FlowElementsContainer
2 inv bp_useExplicitStartAndEndEvents:
3 (bpmnElements()->reject(oclIsTypeOf(BoundaryEvent) or
4 oclIsTypeOf(MessageFlow) or
5 oclIsTypeOf(SequenceFlow))->size() > 0)
6 implies
7 ((self.totalNumberContainerEndEvents() > 0) and
8 (self.totalNumberContainerStartEvents() > 0))
5.2.2.2 Simultaneous merging and splitting gateway should be avoided
A gateway with several inputs and several outputs at the same time may cause misun-derstandings and should be avoided [All10].
The best-practice rule regarding gateway can be enforced by attaching the followinginvariant to the FlowElementsContainer element of the BPMN metamodel.
Listing 5.5: Simultaneous merging and splitting gateway should be avoided.1 context FlowElementsContainer
5. VERIFICATION OF BPMN MODELS 5.2. BPMN Rules Formalization
Figure 5.5: Simultaneous merging and splitting gateway should be avoided
Use: Gateway2 and Gateway2a with distinct merging and splitting roles (top).
Avoid: Gateway2 with two simultaneous merging and splitting roles (bottom).
5.2.2.3 An event should have at most one outgoing Sequence Flow
A Gateway or Activity should be used if a split is needed after an event (not EndEvent),in order that only one outgoing sequence flow comes out of that event [All10].
The best-practice rule regarding events can be enforced by attaching the followinginvariant to the Event element of the BPMN metamodel.
Listing 5.6: An event should have at most one outgoing Sequence Flow.1 context Event
Figure 5.6: An event should have at most one outgoing Sequence Flow
Use: Since IntermediateEvent1 is a split event, a Gateway follows it before the split (top).
Avoid: IntermediateEvent1 is a split event (bottom).
5.3 Conclusion
When a metamodel expresses the abstract syntax of a language, such as in the case ofBPMN metamodel, OCL clauses can be used in a declarative way, similar to 1st orderpredicate logic, to strengthen the semantics of the language, namely by imposing well-formedness rules that reduce the sources of modeling malformation. This chapter pro-posed the formalization and implementation of BPMN rules, informally specified in theBPMN standard, by adding preciseness to the OMG’s BPMN metamodel. This chaptercontributed to enhance the correctness of produced process models, through the deriva-tion of a set of well-formedness rules (section 5.2.1) and best-practices design rules (sec-tion 5.2.2) that BPMN models must attain. Since the rules are embedded in the BPMNmetamodel, process models’ correctness became intrinsically verified by the languageand not ensured by rules implemented in other languages, external tools or checkers.
91
[This page is intentionally blank]
6Measurement of BPMN Models
"One thing I have learned in a long life: that all our science, measured against reality,is primitive and childlike – and yet it is the most precious thing we have."
Context: Nowadays organizations are process-intensive and collaborative, with mea-surement and assessment of process models having a variety of applications includingprocess models’ quality prediction, process improvement and task planning.Objective: Derive a set of grounded BPMN measures that could help process modelersassess BPMN models quality in design time.Method: The chapter set up a BPMN measurement terminology and provides a frame-work for BPMN measurement instantiation. Then, the framework is used for derivationof BPMN measures.Results: It was derived a set of theoretical validated base measures for the internalattributes of BPMN models.Limitations: The proposed framework needs to be instantiated with other measures inorder to assess the suitability of the proposed steps for measures derivation and valida-tion.
93
6. MEASUREMENT OF BPMN MODELS 6.1. Introduction
Conclusion: The set of proposed base measures for BPMN models, after being subjectto a theoretical validation, is available for being subject to an empirical validation.
6.1 Introduction
Process models measurement play an important role in the assessment and improvementof the quality of process modeling. Nowadays, in process-intensive and collaborative or-ganizations, the measurement and assessment of business process diagrams have a vari-ety of applications including process models’ quality prediction, business improvement,and task planning [MSW10].
Measurement is a mechanism that can help answering several questions concernedwith the activity of modeling processes:
• It helps, during the course of process design, to assess its modeling progress, to takecorrective actions based on the assessment of process models’ internal characteris-tics (e.g. process model size, entanglement, autonomy, etc.);• It allows assessing the strengths and weaknesses of the current process models (e.g.,
the frequency and density of certain types of errors);• It provides a rationale for adopting best-practice techniques (e.g., measure the im-
pact of a certain technique on preventing errors in process models).
Being the aim of BPMN to provide a standard language for process modeling, thestandardization efforts can naturally be taken one step further in order to provide BPMNquality measures. This would certainly contribute to define the thresholds that goodprocess models would aim to comply with. Furthermore, it would also facilitate thebenchmarking among organizations’ process models.
If a measure is going to be proposed for BPMN process modeling, the researchershould demonstrate to the community whether it is actually representative of the at-tribute intended to be characterized. Moreover, it is required to scrutinize the measurethrough a methodological approach [BEEM95].
A two steps measurement validation process, is usually sought as the way of ensuringthat a measure is accepted by the corresponding community. Through theoretical valida-tion the researcher should use logic to formally verify whether a measure is meaningfulor not. Empirical validation, on the other hand, implies using data collected from observa-tion or derived by experimentation to support the relevancy of proposed BPMN measure[MSW10]. As is generally recognized (e.g. [BEEM95, MSW10]), both theoretical andempirical validations of measures are necessary and complementary. In this chapter wefocus mainly the theoretical validation of BPMN measures. In chapter 9 will be addressedthe related counterpart of empirical validation of proposed measures, by describing theresults attained from data collected from archival research.
Our intention in this chapter is not to build, from scratch, a methodology for BPMNmeasures’ validation. Our approach consists in the adaption to the context of BPMN of an
94
6. MEASUREMENT OF BPMN MODELS 6.2. Terminology on Process Modeling Measurement
existent framework for measures’ validation from another domain where the measure-ment process is more mature: the Software Engineering field [BEEM95]. The frameworkused for building the BPMN measures is adapted and instantiated from the measuredefinition process Goal Question Metric/MEtric DEfinition Approach (GQM/MEDEA)[BMB02], a systematic approach proposed for software measurement, with a set of guide-lines for the design and definition of sound software measures. The GQM/MEDEAframework received several contributions and influences from the measurement litera-ture, among them, from Goal Question Metric (GQM) paradigm [BCR94] regarding thegoals and models definition through a top-down approach.
The BPMN measurement framework which is presented in the following sections,is intended to define new BPMN measures preventing possible sources of threats thatcould hamper the validity of those measures. Furthermore, given the little value addedacknowledged by the new measures per se [BEEM95], it is intended to go further andcontribute to solve practical problems, namely using the new measures to explain somequality aspects found in actual BPMN process models.
To sum up, this chapter, besides providing some grounded quality measures forBPMN models (section 6.4), intends also to contribute for setting up a BPMN measure-ment terminology (section 6.2), as well as the derivation of a customized framework forBPMN measurement (section 6.3), before the conclusion (section 6.5).
6.2 Terminology on Process Modeling Measurement
In fields such as Software Engineering there are no unequivocal consensus about manyof the concepts and terminology regarding the measurement process [GBC+06]. Nowonder that the similar problem has been faced when trying to collect the set of termsand vocabulary regarding process modeling measures, a relatively recent and immaturefield.
The vocabulary concerning the measurement of quality characteristics is sometimesconflicting and inconsistent among the several sources and references that can be used byresearchers and practitioners. So, we felt the need to set up terms and concepts, aimingto contribute to the harmonization of the BPMN measurement terminology, even beforeproposing a customized BPMN measurement framework and a set of BPMN measures.The set up of the terms used herein, was based on concepts and definitions of measure-ment in the discipline of Software Engineering, as well as from the metrology vocabulary,by closely following the approach of SMO (Software Measurement Ontology) by Garciaet al. [GBC+06].
Several contributions have enriched, during the last decades, the measurement termi-nology particularly in the Software Engineering domain. It was headed an initial effortthrough the CMM (Capability Maturity Model for Software) [PCCW93], later super-seded by the CMMI (Capability Maturity Model Integration) [SEI10a, SEI10b, SEI10c],
95
6. MEASUREMENT OF BPMN MODELS 6.2. Terminology on Process Modeling Measurement
and the international standard ISO/IEC 15504 (Software Process Improvement and Capa-bility dEtermination – SPICE) [ISO04a], derived from process life cycle standard ISO/IEC12207 and from maturity models.
The industry’s concerns about measurement in quality management systems was re-flected in ISO 9001 [ISO08], and also translated to the specific case of computer softwareand related support services, through the ISO/IEC 9000-3 standard [ISO04b], which isfocused in the activities of acquisition, supply, development, operation and maintenancerelated to computer software.
The ISO organization made an effort to harmonize the measurement terminologyby delivering the international vocabulary of metrology (VIM) [ISO07b], covering theterms and concepts related to measurement, later extensively adopted by other stan-dards. Among the international standards that address software measurement conceptsand terminology we highlight some of the major references:
• the IEEE 610.12-1990 [lEE90], a glossary of Software Engineering terminology;• the IEEE 1061-1998 [IEE98], regarding a software quality measures methodology;• the ISO/IEC TR 14143-3 [ISO03] establishes "a framework for verifying the state-
ments of a functional size measurement method and/or for conducting tests re-quested by the verification sponsor, relative to a specified set of performance prop-erties";• the ISO/IEC 25000 series (Software product Quality Requirements and Evaluation –
SQuaRE) [ISO05b] is a set of standards that provides methods for measurement, as-sessment and evaluation of software product quality, and proposes a software prod-uct quality model, and measures for internal quality, external quality, and qualityin use;• the ISO/IEC 15939 [ISO07a] defines the activities of the measurement process for
System and Software Engineering and management disciplines required for ade-quately specify: (a) the measurement information needed, (b) the way measuresand analysis results should be applied, and (c) the way of assessing the validity ofresults analysis;
Relevant proposals regarding software measurement came also from the researchcommunity, namely the set of proposals for defining measures of product attributes inSoftware Engineering [KPF95, BMB02].
From the several efforts due to researchers and practitioners, discrepancies, gaps, andterminology conflicts were revealed [GBC+06]. Terms commonly used (e.g. measure,metric, measurable attribute, measurement) raised the debate and disparate interpreta-tions regarding their precise meaning were suggested. Eventually, terminological agree-ments have emerged in the more recent standards (e.g. ISO/IEC 25000 [ISO05b] andISO/IEC 15939 series [ISO07a]) that chose not to use the more controversial terms, suchas metric, using instead more consensual terms of measurement and measure.
Since this dissertation is about BPMN process modeling, we are naturally concerned
96
6. MEASUREMENT OF BPMN MODELS 6.2. Terminology on Process Modeling Measurement
about terms regarding the definition of concepts required to establish the scope and ob-jectives of the measurement process, as well as those related to the characterization ofBPMN measures. A natural approach to achieve such purposes was by using the generalaccepted measurement vocabulary distilled in the above mentioned references. Never-theless, some alignment on terminology definitions to the process modeling field wasneeded. Hence, the concepts used in this dissertation for now on, regarding BPMN mea-surement, and the relationships among them, is depicted in Figure 6.1, and are based onthe following definitions:
Figure 6.1: Concepts regarding process modeling measurement
Process Model: A set of intertwined process elements depicted in a diagram using aspecific notation (BPMN in our case).
Sub-Model: One of the parts that make up a process model. A sub-model may be alsosubdivided into other sub-models. In BPMN, a sub-model may be embedded inthe model as a regular process diagram depicting an instance of SubProcess or beexternal to the main model as an instance of a CallableElement.
Process Model Element: A metaclass embodying a concept of the process modeling lan-guage which is depicted in process models. BPMN elements that can be seen de-picted in diagrams are shown in Figure 3.4.
Attribute: A particular characteristic of a process model. For example in a BPMN dia-gram, the attribute that denotes the number of elements that constitutes the modelcould be the model’s size. Attributes are involved in empirical hypotheses formula-tion, in the definition of explanatory or outcome variables. (Adapted from ISO/IEC
97
6. MEASUREMENT OF BPMN MODELS 6.2. Terminology on Process Modeling Measurement
25000 series [ISO05b])
Internal attribute: A characteristic that a process model owns by itself and is not depen-dent of any other characteristic. For example, the size of the model does not dependon its entanglement. (Adapted from ISO/IEC 25000 series [ISO05b])
External attribute: A quality characteristic which is dependent on some intrinsic charac-teristics of the model. Appropriate internal characteristics of a process model area pre-requisite for achieving a required external attribute, and appropriate externalattributes are a pre-requisite for achieving quality in use by process modelers. Anexample of external attribute (quality characteristic) of a process model is under-standability that could be dependent upon internal attributes such as size. (Adaptedfrom ISO/IEC 25000 series [ISO05b])
Quality Model: The set of external measurable attributes as well as the relationships be-tween them which provide the basis for specifying quality requirements and eval-uate the quality of BPMN models. The quality of a BPMN model is the degree towhich the model satisfies the stated and implied needs of its various stakeholders(e.g. process analysts, process implementers), and thus provides value. A qual-ity model categorizes model quality into external attributes (quality characteris-tics, which in some cases are further subdivided into sub-characteristics). (Adaptedfrom ISO/IEC 25000 series [ISO05b])
Measure: An abstract value amenable of being materialized by applying a measurementto a particular characteristic (internal or external attribute) of a process model. Ameasure has a predefined measurement scale. (Adapted from ISO/IEC 25000 series[ISO05b])
Base Measure: A measure of an attribute that does not depend upon any other measure.The size measure is a base measure since it only considers elements of the processmodel. In predictive models, base measures are involved as independent variables.(Adapted from ISO/IEC 25000 series [ISO05b])
Indirect Measure: A measure that is a quantification of some quality characteristic. Itis derived from other base or indirect measures. A measure predictor for processmodel correctness is an indirect measure since relies upon other measures. In pre-dictive models indirect measures are involved as dependent variables. (Adaptedfrom ISO/IEC 25000 series [ISO05b])
Measurement Scale: The nature of the relationship between measure observations re-garding the specific sort of mathematical properties (e.g. equality, inequality, ad-dition)(Adapted from ISO/IEC 15939 [ISO07a]). Within measurement theory, afield taken from Mathematics, there are different scales that a measure can take theform of, which provides a set of allowable transformations for that scale. Four main
98
6. MEASUREMENT OF BPMN MODELS 6.2. Terminology on Process Modeling Measurement
measurement scales are often used [FP98] by increasing the degree of accuracy andexpressiveness:
• Nominal (aka categorical variable): if there is no implicit order or distance amongthe categories of a variable, to which the measured observations are assigned.The only transformation admissible to a nominal measure is equality;• Ordinal: if the measured observations are discrete an can be ranked (ordered).
The order of the measured observations matters but not the difference betweenvalues. Besides the equality transformation, inequality using relational opera-tors (e.g. >, <), is also an admissible transformation;• Interval: if the continuous variable can be ranked, as in an ordinal variable, and
additionally the magnitudes of the differences between two values are mean-ingful. Admissible transformations include equality, inequality, addition, andsubtraction.• Ratio: if a continuous variable beside the interval scale features has a meaning-
ful zero value (a beginning or ending point), so that a meaningful rate betweentwo measured observations is allowed. Admissible transformations includeequality, inequality, addition, subtraction, multiplication, and division.
Unit of Measurement: a particular magnitude, defined and adopted by convention withother quantities with the same meaning are compared to express their relative mag-nitude. The number of activities per subprocess could be the unit of measurementof the measure modularization. (Adapted from VIM [ISO07b])
Measurement: A logical sequence of operations aiming to instantiate a particular mea-sure for a given attribute. The resulting instance is called measure observation.(Adapted from VIM [ISO07b])
Measurement Method: A set of operations performed through an algorithm or calcu-lation, aiming to derive a particular base measure for a given internal attribute.(Adapted from ISO/IEC 25000 series [ISO05b])
Measurement Function: A set of operations performed through an algorithm or calcu-lation, by combining base or indirect measures aiming to instantiate a particularindirect measure for a given external attribute. (Adapted from ISO/IEC 25000 se-ries [ISO05b])
Measurement Goal: The goals’ specification intends to provide guidance for the vali-dation process, and the definition of the scope for interpreting the results of thevalidation. The measurement goal supports the definition of targets for base andindirect measures in process modeling. The measurement goals constraint the mea-surements of values of internal or external attributes attained by specific measures.As an example of a measurement goal one can point out the targets for base mea-sures of BPMN models such as size or modularity, assuming they are related with
99
6. MEASUREMENT OF BPMN MODELS 6.3. A Framework for Measurement of BPMN Models
the indirect measure of correctness of BPMN models, resulting from failures in com-plying with the well-formedness rules of the BPMN language [BCR94].
Measure Observation: The quantity or quality assigned as a value to an attribute of aprocess model through a measurement process. (Adapted from ISO/IEC 25000series [ISO05b])
Measure Threshold: a target used as pattern to determine the level of acceptance or re-jection of the measure observed in a certain attribute of a process model. (Adaptedfrom ISO/IEC 15939 [ISO07a])
6.3 A Framework for Measurement of BPMN Models
This framework aims the specification of a set of activities to be developed in order toderive a set of measures for a prediction model in the context of BPMN. Like any method-ological approach, by following and documenting a series of steps, a researcher is ableto trace all design decisions made in the process as well as, justify changes and reviewsmade, as more logical or empirical evidence is unveiled. Besides, it will provide the basisfor replications by other researchers.
6.3.1 Overview
For a certain measure to be generally accepted regarding its interpretation, as well asits usage in the context of the process modeling language such as BPMN, it is necessarythat the measure has been derived through a well-grounded, relevant, meaningful andlogically correct process [MSW12], i.e., the measure should have been validated.
To be carried out the measurement process validation, the language abstract syntax(i.e., the BPMN metamodel) plays an important role, since it contains the elements thatwill be used for the measure specification and implementation. The measure definitionmust be provided in a clear form, backed up by a measurement method, i.e., computedby a deterministic algorithm. This is crucial for measures’ validation and also importantfor the measure’s correct reproduction so that it can be analyzed and studied by otherresearchers and practitioners. To fulfill this requirement, in this work, measures are de-fined and implemented using OCL, which allows building rigorous expressions basedon BPMN models’ elements.
To extract and analyze measure observations from actual BPMN process models,which are the sources of information for the measurement process, the measurementgoals must also be formulated upon the BPMN metamodel.
In this work, we assume that measures’ validation is done through a set of four coarsegrained steps. The process diagram in Figure 6.2 depicts the operational structure of theproposed framework and represents the dynamic view over the concepts and relation-ships of the static model on process modeling measurement depicted in Figure 6.1.
100
6. MEASUREMENT OF BPMN MODELS 6.3. A Framework for Measurement of BPMN Models
Figure 6.2: Measures’ definition and validation
1. BPMN Measurement Inception.The objectives of measures’ designers are summarized into measurement goals,based on their knowledge about the BPMN standard and regular usage of the no-tation producing process models. Driven by measurement goals, a set of intuitive1
hypotheses are established highlighting the qualitative relation between internal at-tributes of process models (independent variables candidates) to other external at-tributes of the same models (dependent variables candidates). For instance, startingwith predicting the correctness of BPMN models as a measurement goal, a mea-sures’ designer could possibly conjecture that could exist a positive association be-tween the BPMN model size and the number of errors found in the model.
2. Definition of Base Measures for Internal Attributes.This phase evaluates the suitability of base measures for conveying the internal at-tributes characteristics of process models. Internal attributes capture factors thatare assumed to have a causal relationship with external attributes. Internal at-tributes are formalized by a set of mathematical properties (e.g. non-negativity,monotonicity). The theoretical validation of a base measure is a formal approach toassess whether the measure is valid or not, regarding the properties of internal at-tributes. The properties should be as generic as possible, i.e., independent of theprocess models’ representation. Underpinned by properties, base measures are de-fined, as well as the measurement method that must be performed over a set of pre-defined BPMN model elements. For instance, size is an internal attribute of a BPMNprocess model whose base measure could be defined by the number of FlowNode el-ements in the model, calculated through a specific measurement method.
3. Definition of Indirect Measures for External Attributes.The external attributes are quality characteristics of BPMN process models (e.g. cor-rectness, understandability, maintainability). Those are the qualities that process
1At this primary stage, we use the term intuitive rather than empirical, reserving the former term for themore advanced stage of the empirical study [Spe81].
101
6. MEASUREMENT OF BPMN MODELS 6.3. A Framework for Measurement of BPMN Models
models should convey when they are in use, namely by process analysts and pro-cess implementers either for documenting sake or for process enactment. Duringthis phase it must be assessed whether the indirect measures adequately capturethe external attributes of the BPMN process model.
4. Measures’ Empirical Validation.The intuitive hypotheses give way to grounded empirical hypotheses, which willbe developed in chapter 9. There, empirical hypotheses are verified using the mea-sures defined for the internal and external attributes. The empirical hypotheses area refinement of the intuitive hypotheses by providing a specific function for the re-lationship between base and indirect measures used to build the predictive model.The adequacy of the predictive model is determined by the level of significance ofresults.
6.3.2 Detailed Activities
6.3.2.1 BPMN Measurement Inception
Measures should be defined and analyzed with respect to a clear set of objectives. It isimportant to establish the purpose of information needed, as well as how the informationshould be collected and analyzed from BPMN models in order to achieve goals.
After goals definition and prior to the precise definition of measures and checking ofits mathematical properties, we must agree about the intrinsic meaning of the measures,as well as formulate statements regarding the related attributes of process models thatwe believe can be theoretically and empirically relevant.
6.3.2.1.1 Define Measurement Goals
There is a need for responses to questions regarding the quality of BPMN models, so thecollected data must be driven by this purpose. Since it is also needed a specification forthe interpretation of the collected data, the derivation of measures should be done fromexplicit goals. So, we need a goal-oriented paradigm.
The Goal Question Metric (GQM) [BCR94] approach is a systematic approach builtwith the aim of filling the gap between goals and measures in a systematic manner bymeans of questions and models. According to GQM the measurement goals should beranked based on, among other things, their perceived relevance to the intermediate goals,and their feasibility regarding the building of a prediction model. GQM provides a tem-plate and guidelines to define measurement goals and refine them into concrete and real-istic questions, which subsequently lead to the definition of measures. According to theGQM goal template, measurement goals have a viewpoint, a purpose, and an environment.
A GQM model is developed here by identifying a set of quality goals regarding BPMNmodels. From those goals and based upon the BPMN metamodel, questions are derivedthat define them as completely as possible. For characterizing the model with respect to
102
6. MEASUREMENT OF BPMN MODELS 6.3. A Framework for Measurement of BPMN Models
a certain set of quality issues (e.g. correctness), a quality model must be derived for deal-ing with those issues (e.g., size, complexity). Questions try to characterize the processmodel to be measured, with respect to a selected quality issue, and from a selected view-point. The next step consists in specifying the measures that are required for answeringthose questions, and to track the conformance to the goals. After the measures have beenspecified, we need to develop the data collection mechanisms, including validation andanalysis mechanisms. A set of data is associated with every question in order to answerit in a quantitative way.
6.3.2.1.2 Define Intuitive Hypotheses
An important step for deriving a valid and useful measure is understanding its intuitivenotion regarding what the attribute is supposed to characterize. Depending on the at-tributes we want to measure, can be used different BPMN constructs (e.g. consider allpossible elements of type FlowNode of Figure 3.7 or disjoint sets of constructs, such asprocessing nodes Activity, triggering/catching Event and control-flow Gateway). Onlyqualitative statements are made in this phase, based on the intuition and practical usageof the attributes. So, the definition of an intuitive measure is inherently subjective and isnot quantifiable.
Intuitive hypotheses help to identify internal attributes that are believed to be rel-evant to the intermediate goals and facilitate the process for defining measures and tointerpret the results. In general, an hypothesis involves several internal attributes andone external attribute. The intuitive hypotheses stated in this section are different fromthose that are defined when measures will be empirically validated through statisticaltest of hypotheses. The main differences are that intuitive hypotheses are:
• defined in terms of attributes (and not measures);• formulated in the terms of a statement that is intended not to be rejected (i.e., the
same as the alternative hypotheses of the regular statistical test).
It is important to ensure an agreement on the basic set of properties that the intuitivemeasurement systems of common internal attributes should have. This will help BPMNresearchers and practitioners to use the same terminology when deriving their measures.
We call upon a definition of intuitive measurement system2 for modeling the intuitionand practical knowledge (adapted from [BEEM95]).
Definition 6.1. Intuitive Measurement System
I is an Intuitive Measurement System, denoted as I = (P, E, R1, · · · , Rn), where P is anonempty set of BPMN models with attributes to be measured. Each P has a set ofE elements which are instances of BPMN metamodel’s of graphical elements (e.g.FlowNode, ItemAwareElement – see Figure 3.4). Ri is the kii-ary intuitive relations onattributes of P with i = 1, · · · , n (e.g., the intuitive relation between a process model
2This concept is named empirical measurement systems in measurement theory [FP98].
103
6. MEASUREMENT OF BPMN MODELS 6.3. A Framework for Measurement of BPMN Models
being greater than another one, or the intuitive closed binary operation of a processmodel being the addition of the size of its sub-models).
The intuitive measurement system describes the reality on which measurement iscarried out (via the BPMN set of elements E of process model P), our common sense andknowledge of the process models’ attributes we want to measure (through the collectionof intuitive relations Ri’s).
6.3.2.2 Definition of Base Measures for Internal Attributes
The measures that appear in the previously defined measurement goals (section 6.3.2.1.1)must be formalized in order to exhibit soundness properties. To describe the character-istics of measures for a given attribute, are used mathematical properties, which ensuresthe measure’s theoretical validity based on measurement theory.
6.3.2.2.1 Formalize Internal AttributesInternal attributes should be defined to support the intuition expressed in 6.3.2.1.2. Theymust possess a set of properties that measures must comply with, for being well for-malized. Well-defined internal attributes are related with the specific characteristics ofBPMN models or sub-models (Definition 6.2) that are intended to be measured. Theproperty-based approach provides an adequate and rigorous characterization of internal at-tributes [BMB96]. So, we will follow this approach to formalize the internal attributes’properties (Definition 6.3) of BPMN models.
Definition 6.2. BPMN Model
P is a set of BPMN models represented as a pair [E, L], where E consists in theset of graphical elements of P, and L is a set of binary relations on E (L ⊆ E ×E) representing the set of graphical relationships defined in the BPMN metamodelbetween elements of P (e.g. SequenceFlow, MessageFlow, DataAssociation – see Figure3.4).
Given a set of BPMN models P=[E, L], m=[Em,Lm], where Em consists in the subsetof elements of P. m is a sub-model of P if and only if Em ⊆ E, Lm ⊆ Em × Em, and Lm ⊆L. This is denoted by m ⊆ P. Furthermore, ∀ e ∈ E (∃ m ∈ P (m=[Em,Lm] ∧ e ∈ Em))∧ ∀ l ∈ L (∃m ∈ P (m=[Em,Lm] ∧ l ∈ Lm))
Note: It is not imposed that for two sub-models m1 = [Em1, Lm1] and m2 = [Em2,Lm2], Em1 ∩ Em2 = ∅ (see Figure 6.3 where DataObject1 is shared by two sub-models)
Definition 6.3. BPMN Model Internal Attribute
An internal attribute A is a characteristic of a BPMN model, which owns a set ofmathematical properties α1, . . . , αi that must be complied by the measure of A.
104
6. MEASUREMENT OF BPMN MODELS 6.3. A Framework for Measurement of BPMN Models
Figure 6.3: A BPMN Model P constituted by 3 sub-models: a generic sub-model (top),the sub-model of SubProcess1 (middle) and the sub-model of SubProcess2 (bottom).
6.3.2.2.2 Define Base MeasuresFor the theoretical validation that we want to be carried out, we need as part of themeasurement theory, to map the intuitive relations in the intuitive measurement system(see Definition 6.1), onto relations between the values of measures, which occurs in a socalled formal measurement system, formally defined as (adapted from [BEEM95]):
Definition 6.4. Formal Measurement System
F is a Formal Measurement System, denoted as F = (Q, S1, · · · , Sn), where Q is a non-empty set of formal entities (e.g. numbers or vectors) and Si are ki-ary relations onQ (e.g. ">" or "+").
The formal measurement system describes (via the set Q) the values of the measuresof process models’ attributes. The values may be integer numbers, real numbers, vectorsof integer and/or real numbers. A formal measurement system also describes (via thecollection of relations Si’s) the relations of interest for the measures.
105
6. MEASUREMENT OF BPMN MODELS 6.3. A Framework for Measurement of BPMN Models
Every element p ∈ P, in the intuitive measurement system, is mapped into a valueof Q, i.e., after being subjected to a measuring process it is mapped to a measure υ(p).Every intuitive relation Ri is mapped into a formal relation Si. For instance, the relationgreater than between two BPMN models is mapped into the relation ">" between thesize measures of those models. The formal relations must preserve the meaning of theintuitive statements. For instance, suppose that R1 is the intuitive relation greater than,and S1 is the formal relation ">", υ is a size measure if and only if for any two processmodels P1, P2, with process model P1 greater than process model P2, we have υ(P1) >υ(P2).
A measure for an internal attribute can be defined as:
Definition 6.5. BPMN Model Measure
A measure υ conveys the valuation of an attribute A in a BPMN model P, denotedby PA, such that υ:PA→Q yields a formal entity υ(PA) ∈Q, which after instantiationbecomes a measure observation.
This leads to the following definition of a measurement scale:
Definition 6.6. Measurement Scale
Let I = (P, E, R1, · · · , Rn) be an intuitive measurement system, F = (Q, S1, · · · , Sn)a formal measurement system and υ a measure. The triple (P, Q, υ) is a scale ofmeasurement if and only if for all i, j and for all a1 , · · · , ak, b, c ∈ P the followingholds Ri (a1, · · · , ak)↔ Si (υ(a1),· · · ,υ(ak)) and υ(b R0 c) = υ(b) S0 υ(c). If Q =R, withR as the set of real numbers, the triple (P, Q, υ) is a ratio scale.
The properties for the measures of process models’ attributes, besides being logicallyconsistent, should also hold for the admissible transformations of measures’ measure-ment scales. The choice of the scale of measurement must be based on the precision ofthe measurement goals and results usage.
The measures we want to define are intended to measure standard attributes of pro-cess models, such as size, autonomy, or entanglement. We must ensure that what themeasure conveys is actually what we look forward. For instance, we must check whethera measure we defined to measure the size of a process model is really a size measure. Thiscan be done by using a set of properties which characterize the measures of a specified at-tribute. Sets of properties, are designed to facilitate such a verification procedure. Theseproperties must be related with the attributes’ intuition. Instances of such properties canalso be collected from the literature (e.g. [Wey88, BMB96]).
Since those properties are grounded in measurement theory, they allow an initial for-mal perspective and definition based on the mathematical properties of measures, beforeis conducted a experimental validation. Measurement theory is a useful formal tool forbuilding and validating measures and hence ensuring, through theoretical validation,whether the measure actually quantifies a concept intended to assess. This means that
106
6. MEASUREMENT OF BPMN MODELS 6.3. A Framework for Measurement of BPMN Models
a measure υ of an attribute must be consistent with the intuitive understanding of thisattribute. Theoretical validation supports modeling the intuitive understanding of theattribute to be measured. For instance, if υ is supposed to measure the understandabilityof process models, being P1 and P2 two process models, it is required that one can ensurethat υ(P1) > υ(P2), before empirical studies could substantiate that process model P1 ismore easy to understand than process model P2.
The abstraction chosen for capturing relationships among process models’ elementswas the BPMN metamodel. BPMN constructs are the components which can be used toinstantiate an actual BPMN process model. Such an instance is expressed graphically asa directed graph, consisting of nodes (e.g instances of FlowNode) and links (e.g. instancesof SequenceFlow) [Fra03]. This abstraction helps expressing the intuition regarding theinternal attributes’ properties and substantiate the base measures definition.
6.3.2.3 Definition of Indirect Measures for External Attributes
The definition of measures for the external attributes mirrors the steps shown in section6.3.2.2 regarding internal attributes. To describe the characteristics of indirect measuresfor a given external attribute, are used statistical properties, which ensure the measure’stheoretical validity based on Statistical Theory.
6.3.2.3.1 Formalize External AttributesExternal attributes are more tangible than internal attributes. For instance, a BPMNmodel’s external attribute such as understandability is much more easy to understand,on an intuitive level, than the complexity BPMN model’s internal attribute. Researchersand practitioners are more acquainted with external attributes, therefore the need for for-malizing the properties of their corresponding measures is seen as less important thanthat for formalizing the properties of measures for internal attributes.
There are external attributes of process models that can be interpreted using the prob-ability properties. An axiomatic was given to the concept of probability in the Theory ofProbability, a domain widely used, for example, to draw in formal terms inferences aboutthe expected frequency of events. Since this knowledge on probability is already avail-able, one can apply it to the current BPMN measurement context. In the theory of proba-bility the representation of probabilistic concepts is given in terms that can be consideredseparately from their meaning. These terms are manipulated by the rules of mathemat-ics and logic, and any results can be interpreted and translated back into the problemdomain. From the theoretical attempts to formalize probability, we use the Kolmogorovformulation. In Kolmogorov’s perspective, sets are interpreted as events and the proba-bility itself as a measure on a class of sets [DS02].
Kolmogorov introduced the notion of probability space, a mathematical construct witha specific kind of situation or experiment in mind, which models a real-world processconsisting of states that occur randomly [AL06]. The probability space consists of threeparts:
107
6. MEASUREMENT OF BPMN MODELS 6.4. Measures Derivation for BPMN Process Models
• A sample space, Ω, the set of all possible outcomes.• A set of events F, where each event is a set containing zero or more outcomes.• The function P from events to probabilities, i.e., the assignment of probabilities to the
events.This notion of probability space, together with other axioms of probability introduced
by Kolmogorov, are the foundation for the indirect measures derivation in this work.
6.3.2.3.2 Define Indirect MeasuresThe validation of base measures was shown consisting mainly in checking the measures’compliance with a set of properties. This may not always be the case for indirect mea-sures since they usually cannot be precisely characterized [BMB96]. However, when us-ing statistical tools, it is still possible to use theoretical techniques for validating indirectmeasures’ properties. For instance, it can be mathematically demonstrated that statis-tical tools that are intended to be used, such as correlation [Was04] and Binary LogisticRegression (BLR) [CC97], satisfy the probability properties [RPD98, MR03].
6.4 Measures Derivation for BPMN Process Models
Since a BPMN model is made up by the composition of many different elements (activi-ties, data repositories, control-flows), which can be analyzed from different perspectives,there is no unique concept, which measurement could act as a universally accepted mea-sure for BPMN models’ characteristics. Hence, we need to call upon a set of measures forevaluating the different internal attributes of process models. We want those measures tobe associated with BPMN models’ quality characteristics (e.g. correctness) and provideguidelines and hints for attaining better process modeling artifacts.
It is expected that in the future, as happened with modeling and development tools inSoftware Engineering, BPMN modeling tools would be able to provide process modelingmeasures to aid process analysts and designers to build process models, beyond the merecounting of elements found as indicators in current tools. There are already some pro-posals regarding BPMN measures, however with the limitations pointed out in section4.4.
To ensure the soundness of our measures’ proposals they were validated according tothe framework detailed in section 6.3. As already mentioned, from the overall validationprocess (see Figure 6.2), in this section we are concerned, with theoretical validationleaving empirical validation for chapter 9.
6.4.1 BPMN Measurement Inception
6.4.1.1 Define Measurement Goals
BPMN models’ quality is the focus of this dissertation. Therefore, the highest level, con-cerning goals, was assigned to the aim of attaining correct BPMN models. This objective
108
6. MEASUREMENT OF BPMN MODELS 6.4. Measures Derivation for BPMN Process Models
is the most important one because faults in BPMN models can have impact in all businessprocess’s life cycle. Since this is a very broad goal, we have to decompose it in interme-diate goals. There are several BPMN model attributes (e.g. size, modularity) that areworth studying in our dissertation, since we believe that each one has some sort of in-fluence in BPMN models’ correctness and therefore BPMN models’ quality. Studying theBPMN models’ attributes is feasible with the available resources and constraints. Study-ing other aspects (e.g. process modelers’ experience) that we presume are also influentialfor BPMN models’ quality would require other resources and case studies that are diffi-cult to obtain, so that will not be undertaken in this dissertation.
To fill the GQM goals’ template, we have to define relevant abstractions for BPMNmodels’ attributes. We have also to collect enough data about the quality focus to allowfor a statistically significant validation of the relationships between the attributes’ mea-sures of BPMN models and the quality characteristic we are focused. The aim is thatmeasures can be used for defining prediction systems. Furthermore, correctness is thequality focus (external attribute) and it will be assessed by checking the conformance ofBPMN models with BPMN well-formedness rules derived in section 5.2. We are con-cerned with violations of those rules in BPMN models specified in the BPMN standard[BPM11]. Those faults are relevant from the viewpoint of either process analysts andimplementers. The non conformance of BPMN best-practices rules by process modelersare not our concern in this section. The source for our data collection is a set of openrepositories of BPMN model snippets.
Taking the GQM template, we formulated the top-level goal GB in the followingterms:
Analyze BPMN modelsfor the purpose of predictionwith respect to assessment of quality characteristic of correctness,from the point of view of process modelers,in the context of BPMN models collected from open repositoriesconstrained by the requirements of regression studies.
This top-level goal, stated to drive the measures’ theoretical validation through thissection is in line with the research question RQB, formulated in general terms on chapter1. The top-level goal GB is refined next with intermediate goals GB1 and GB2. We decom-posed also GB1 into the sub-goals, GB1a–GB1e, formulating these sub-goals in terms of thedifferences regarding the GB1 statement, as well as the goal related question.
• GB1 – analyze BPMN models for the purpose of measurement models’ internal at-tributes with respect to assessment of quality characteristic of correctness, from thepoint of view of process modelers, in the context of samples of models collectedfrom open repositories constrained by the requirements of correlational studies.Question: How to assess the internal attributes of BPMN models?∗ GB1a – . . . for the purpose of assessing models’ entanglement . . .
Question: How to measure the tangle of BPMN models?
6. MEASUREMENT OF BPMN MODELS 6.4. Measures Derivation for BPMN Process Models
∗ GB1b – . . . for the purpose of measurement models’ autonomy . . .Question: How to measure the autonomy of BPMN models?
∗ GB1c – . . . for the purpose of measurement models’ complexity . . .Question: How to measure the complexity of BPMN models?
∗ GB1d – . . . for the purpose of measurement models’ modularity . . .Question: How to measure the modularity of BPMN models?
∗ GB1e – . . . for the purpose of measurement models’ size . . .Question: How to measure the size of BPMN models?
• GB2 – analyze BPMN models for the purpose of evaluating a prediction model withrespect to assessment of quality characteristic of correctness, from the point ofview of process modelers, in the context of samples of models collected from openrepositories, constrained by the requirements of regression studies.Question: How to evaluate a prediction model of BPMN models’ quality character-istic of correctness?
6.4.1.2 Define Intuitive Hypotheses
Here we state hypotheses on aspects of the BPMN models that are relevant to the measur-able goals. An intuitive hypothesis is a statement believed to be true about the relationshipbetween one or more attributes of the BPMN models (object of study) and the externalattribute correctness (quality focus). The hypotheses formulated below intend to revealour intuitive understanding of the phenomena in study.
As mentioned before, the intuitive hypotheses are candidate hypotheses to the statis-tical tests of hypotheses to be conducted later in the empirical studies (chapter 9). Thelatter should be more precisely formulated when additional information becomes avail-able about measures for the internal and external attributes.
The process models’ measures that we propose are next summarized and their defini-tion adapted to BPMN process modeling, in order to be implemented through the BPMNmetamodel:
1. IHB1a (Tangle and Correctness): High entanglement among instances of FlowNodewill result in a high interrelated model, which is undesirable since changes in onepart of the model could have collateral effects in the interlinked instances. On theother hand, a model with highly interconnected FlowNode instances would have agreater difficult to be understandable or more prone to modeling errors. An in-crease in entanglement should imply an increase in error proneness of the overallBPMN model.
2. IHB1b (Autonomy and Correctness): The better the process modeler is able to en-capsulate related modeling features together, the more reliable and maintainablewill be the designed business process. A more cohesive and autonomous instanceof Activity should encapsulate aspects regarding the responsibility it is supposed toensure. An increase in the autonomy should imply a decrease in error probability
110
6. MEASUREMENT OF BPMN MODELS 6.4. Measures Derivation for BPMN Process Models
of the overall process model. So, high autonomy values are related to lower error-proneness, due to the fact that the changes required by a change in a sub-model areconfined in a well-encapsulated part of the overall model.
3. IHB1c (Complexity and Correctness): Simpler models are preferable to complicatedones, since those are more difficult to verify and more prone to reduce the under-standability quality attribute. An increase in the complexity should imply an in-crease in error proneness of the overall BPMN model.
4. IHB1d (Modularity and Correctness): A low value of modularity means a high levelof flatness, and therefore the model was built in extension and not using instancesof SubProcess, which could affect the BPMN model quality. An increase in the mod-ularity should imply a decrease in error proneness of the overall BPMN model.
5. IHB1e (Size and Correctness): the size of the BPMN model, is measured by theweighted number of instances of Activity, Event and ItemAwareElement includedin the BPMN model. An increase in the size should imply an increase in errorproneness of the overall BPMN model.
6. IHB2 (Prediction Model for Correctness): the internal attributes of the BPMN model,determine its structural characteristics. A prediction model can establish the corre-lation among the internal attributes of the BPMN model and its error proneness.
6.4.2 Definition of Base Measures for Internal Attributes
6.4.2.1 Formalize Internal Attributes
The properties we define here for BPMN models internal attributes are a generalization ofthe properties that authors have already provided in the literature (e.g. [Wey88, BMB96])for Software Engineering. Although there might be some consensus on some of the prop-erties, the acceptance of a set of properties for an internal attribute is ultimately a subjec-tive matter, as well as for any formalization of an informal concept.
We do not claim to address all types of possible properties for each internal attribute.On the other hand, the fact that similar sets of properties are associated with differentinternal attributes is not contradictory, since each property is interpreted differently inthe context of each attribute.
The properties of each of the BPMN models’ internal attributes are presented next:
Tangle – The internal attribute of tangle captures the amount of relationships (instancesof SequenceFlow) between FlowNode elements in a BPMN model. Given a FlowNode in-stance, two kinds of entanglement exists. The inbound entanglement captures the amountof relationships from elements outside the FlowNode instance to the instance itself; theoutbound entanglement captures the amount of relationships from the FlowNode instanceto elements outside the processing element.
Definition 6.7. Tangle – The entanglement of a process model P or a sub-model is afunction Tangle(P) that is characterized by the following properties:
111
6. MEASUREMENT OF BPMN MODELS 6.4. Measures Derivation for BPMN Process Models
Property 6.7.1. Tangle Nonnegativity – The entanglement of a process model P = [E, L] isnon-negative
Tangle(P) ≥ 0
Property 6.7.2. Tangle Null Value – The entanglement of a process model P = [E, L] isnull if, for all processing elements (Ea), the inbound relationships (Lai) and outboundrelationships (Lao) are empty sets
∀ Ea (Lai = ∅ ∧ Lao = ∅)⇒ (Tangle(P) = 0)
Property 6.7.3. Tangle Monotonicity – Given a process model P1 = [E1, L1] and anotherprocess model P2 = [E2, L2], obtained from P1 by adding a set of inbound (L2i) and/oroutbound (L2o) relationships linking to/from processing elements E1a from P1, with L2i
⊆ L2 ∧ L2o ⊆ L2, then the entanglement do not decreases
Tangle(P2) ≥ Tangle(P1)
Property 6.7.4. Tangle Abstracting Sub-Models – Given a process model P1 = [E1, L1] andanother process model P2 = [E2, L2], obtained from P1 by removing the sub-model m1 =
[Em1, Lm1], then the entanglement increases
Tangle(P2) ≥ Tangle(P1)
Autonomy – The concept of autonomy assesses the tightness with which related pro-cessing element (instance of Activity) features are grouped together in a process modelor sub-model. It is assumed that the better a processing element is able to encapsulaterelated business process features together, the more reliable and maintainable the processmodel will be.
Definition 6.8. Autonomy – The autonomy of a process model P or a sub-model is afunction Autonomy(P) that is characterized by the following properties:
Property 6.8.1. Autonomy Nonnegativity – The autonomy of a process model P = [E, L] isnon-negative
Autonomy(P) ≥ 0
Property 6.8.2. Autonomy Null Value – The autonomy of a process model P = [E, L] is nullif processing elements (Ea) or the relationships (L) between the former are empty sets
(Ea = ∅) ∨ (L = ∅)⇒ (Autonomy(P) = 0)
Property 6.8.3. Autonomy Non-Monotonicity – Given a process model P1 = [E1, L1] andanother process model P2 = [E2, L2] with the same set of processing elements E1 = E2,adding relationships between processing elements to the process model P2, such that L1
⊆ L2, do not decrease the autonomy of P2, i.e.,
112
6. MEASUREMENT OF BPMN MODELS 6.4. Measures Derivation for BPMN Process Models
Autonomy(P1) ≥ Autonomy(P2)
Property 6.8.4. Sub-Models Autonomy – Given a process model P1 = [E1, L1] and anotherprocess model P2 = [E2, L2], obtained from P1 by converting interrelated processing ele-ments E1i from P1 in a sub-model m2 = [Em2, Lm2], such that E1i ⊆ Em2 ∧ Lm2 ⊆ L2, thenthe autonomy do not decrease
Autonomy(P2) ≥ Autonomy(P1)
Complexity – The more control-flow edges Lc exists between the elements of a processmodel, the more complex is the process model. Intuitively, one could expect that thecomplexity of a process model was not lesser than the sum of complexities of the sub-models considered per se.
Definition 6.9. Complexity – The complexity of a process model P is a function Com-plexity(P) that is characterized by the following properties:
Property 6.9.1. Complexity Nonnegativity – The complexity of a process model P = [E, L]is non-negative
Complexity(P) ≥ 0
Property 6.9.2. Complexity Null Value – The complexity of a process model P = [E, L] isnull if Lc is an empty set
(Lc = ∅)⇒ (Complexity(P) = 0)
Property 6.9.3. Complexity Monotonicity – The complexity of a process model P = [E, L] isnot lesser than the sum of the complexities of any set of its sub-models m1 = [Em1, Lm1]and m2 = [Em2, Lm2]. None two sub-models share control-flow edges, so any control-flowedge in the process model P (Lc) belongs to one of the sub-models m1 (Lm1c) or m2 (Lm2c)or the process model, and the sub-models do not share control-flow edges.
Property 6.9.4. Control-Flow Relationship Monotonicity – Adding control-flow edges be-tween elements of a process model P1, such that L1c ⊆ L2c, do not decrease the complexityof P1
Modularity – The concept of modularity is related to the decomposition of the mainprocess model in sub-models. Intuitively, one could expected that an high level of mod-ularization in models with less number of graphical elements could enhance the overallmodel.
113
6. MEASUREMENT OF BPMN MODELS 6.4. Measures Derivation for BPMN Process Models
Definition 6.10. Modularity – A measure of the graphical elements that are part of sub-models in a process model, P = [E, L], is a function Modularity(P) that holds the followingproperties:
Property 6.10.1. Modularity Nonnegativity – Modularity cannot be negative
Modularity(P) ≥ 0
Property 6.10.2. Modularity Null Value – Modularity is null when a process model con-tains a unique sub-model m = [Em, Lm] that depicts the overall business process, withoutconsidering the components’ graphical details.
(|m| = 1)⇒ (Modularity(P) = 0)
Property 6.10.3. Modularity Monotonicity – Adding a sub-model mi to a process model,for the same number of processing entities (Ea), cannot decrease the modularity
Property 6.10.4. Modularity Decreasing by Merging Sub-Models – Given a process model P1
= [E1, L1], merging the sub-models of P1, m1 = [Em1, Lm1] and m2 = [Em2, Lm2] originatingthe process model P2 = [E2, L2] does not increase the modularity of the latter
Size – The concept size is informally related to the number of graphical elements that filla process model.
Definition 6.11. Size – A measure of the graphical elements that are part of a processmodel, P = [E, L], is a function Size(P) that holds the following properties:
Property 6.11.1. Size Nonnegativity – Size cannot be negative
Size(P) ≥ 0
Property 6.11.2. Size Null Value – Size is null when a process model does not contain anyelements
(E = ∅)⇒ (Size(P) = 0)
Property 6.11.3. Size Monotonicity – Adding elements to a process model cannot decreaseits size
Corollary 6.11.1. Size Decreasing by Merging Sub-Models – Merging sub-models m1 = [Em1,Lm1] and m2 = [Em2, Lm2] can decrease the size of the process model P
(m1 ⊆ P ∧m2 ⊆ P ∧ E ⊆ Em1 ∪ Em2)⇒ Size(P) ≤ Size(m1) + Size(m2)
114
6. MEASUREMENT OF BPMN MODELS 6.4. Measures Derivation for BPMN Process Models
6.4.2.2 Define Base Measures
For each one of the attributes of BPMN models, base measures are defined by usingabstractions of elements and relationships. The measures are normalized in order toallow meaningful comparisons between different BPMN models. After is defined eachbase measure, it is checked whether the measure is compliant with previously definedproperties of internal attributes they intend to measure.
Definition 6.12. Measure Tangle – We define the measure tangle (tg) of a BPMN modelas the ratio between the total number of edges e (SequenceFlow) over the total number ofnodes n (FlowNode) of a BPMN model.
tg =|e||n|
(6.1)
• Property Tangle Nonnegativity – The property holds if a BPMN model has more thanone FlowNode.• Property Tangle Null Value – The tangle of a BPMN model is null if for all FlowNode
there are no inbound or outbound SequenceFlow.• Property Tangle Monotonicity – Given a BPMN model adding a set of inbound and/or
outbound SequenceFlow will increase the entanglement of the model.• Property Tangle Abstracting Sub-Models – Given a BPMN model and another BPMN
model derived from the first by replacing a SubProcess by its elements, the entan-glement will increase.
Definition 6.13. Measure Autonomy – We define autonomy (aut) of a BPMN model asthe total number of processing nodes a (Activity) over the total number of edges e (Se-quenceFlow).
aut =|a||e|
(6.2)
• Property Autonomy Nonnegativity – The property holds if there is at least a Sequence-Flow instance.• Property Autonomy Null Value – The autonomy of a BPMN model is null if there are
no Activity nodes or SequenceFlow edges.• Property Autonomy Non-Monotonicity – Given a BPMN model and another BPMN
model derived from the first one by adding SequenceFlow edges among Activitynodes, the autonomy will not decrease.• Property Sub-Models Autonomy – Given a BPMN model and another BPMN model
derived from the first one by converting interrelated Activity nodes into a SubPro-cess. The autonomy of the second BPMN model will not decrease.
Definition 6.14. Measure Complexity – We define complexity (cx) of a BPMN model(slightly based on CFC [CMNR06]) as the possible number of mental states [Mil56],which are given by the weighted sum of control-flow nodes (gi) (Gateway), where i refers
115
6. MEASUREMENT OF BPMN MODELS 6.4. Measures Derivation for BPMN Process Models
the OR-splits nodes (o), which receive a weight of 2 for being more difficult to mentallyfollow; parallel-splits (p) that receive the weight 0.5 for being more easily processed; fi-nally the XOR-splits nodes (x) receive a neutral weight of 1. The logarithmic scale wasused to reduce the data range variation.
• Property Complexity Nonnegativity – The property holds independently of the num-ber of Gateway nodes.• Property Complexity Null Value – The complexity of a BPMN model is null if there
are no Gateway nodes resulting from OR-splits, parallel-splits and XOR-splits nodes.• Property Complexity Monotonicity – Given a BPMN model composed by at least two
SubProcess. The complexity of the BPMN model will not decrease regarding thesum of the complexities of the two SubProcess.• Property Control-Flow Relationship Monotonicity – Given a BPMN model and another
BPMN model derived from the first one by adding Gateway nodes resulting fromOR-splits, parallel-splits or XOR-splits nodes. The complexity of the second BPMNmodel will not decrease.
Definition 6.15. Measure Modularity – A low value of modularity mean a high levelof flatness (the Activitiy nodes denoted by a are mainly Task), and therefore the BPMNmodel was built in extension and not by modules (Subprocess or CallActivity denoted bysp).
md =|sp||a|
(6.4)
• Property Modularity Nonnegativity – The property holds if there is at least an Activitynode.• Property Modularity Null Value – The modularity of a BPMN model is null if there
are no SubProcess nodes or only Task nodes.• Property Modularity Monotonicity – Adding a SubProcess to a BPMN model, for
the same number of Activity nodes, cannot decrease the modularity of the BPMNmodel.• Property Modularity Decreasing by Merging Sub-Models – Given a BPMN model.
Merging several SubProcess nodes into one SubProcess does not increase the modu-larity of the BPMN model.
Definition 6.16. Measure Size – The elements that compose size (sz) are weighted duetheir relative importance in the process model from 2, for activities (a), to 0.5 for dataelements (d). Events (v) are given the weight of 1. Gateways are not considered in thesize measure since they were already considered as part of the complexity measure. Thelogarithmic scale was used to reduce the data range variation.
sz = log2(1 + 2× |a|+ |v|+ 0.5× |d|) (6.5)
116
6. MEASUREMENT OF BPMN MODELS 6.4. Measures Derivation for BPMN Process Models
• Property Size Nonnegativity – The property holds if there is at least an Activity, Event,or ItemAwareElement nodes.• Property Size Null Value – The size of a BPMN model is null if there are no Activity,
Event or ItemAwareElement nodes.• Property Size Monotonicity – Adding Activity, Event or ItemAwareElement nodes to a
BPMN model cannot decrease its size.• Corollary Size Decreasing by Merging Sub-Models – Merging SubProcess nodes can
decrease the size of the BPMN model.We have been using the BPMN metamodel for representing BPMN diagrams, which
is a kind of directed graphs where the nodes are instances of type FlowNode and the edges– linking a source FlowNode to a target FlowNode – are instances of type SequenceFlow. Asalready done regarding BPMN validation (section 5.2), we use here OCL for instantiationof the BPMN measures previously defined and theoretically validated. In Listing 6.1 isshown the example of the implementation, as a OCL operation, of the BPMN measuretangle. This approach improves the BPMN metamodel with functionalities regardingassessment of internal attributes of BPMN models. An advantage of our approach is tocontribute for overcoming a common shortcoming of informal definition of measures,which often drives to a different and inconsistent implementation by tool makers.
As will be referred in chapter 7 our BPMN measures were implemented in a UMLenvironment with OCL evaluation support (USE [GBR07]), which allowed the testingand collection of data of BPMN measures for empirical validation (chapter 9).
Our approach regarding measures definition, follows other ones undertaken in therealm of Software Engineering, such as the ones using the Z notation [MM97b, MM97a]and M2DM approach [BeA01].
Listing 6.1: Implementation of tangle measure in OCL
1 abstract class FlowElementsContainer < BaseElement
2 ...
3 operations
4 ...
5 -- returns the measure observation for entanglement
6 tangle(): Real = self.edges() / self.nodes()
7
8 -- returns the number of SequenceFlow in the container and sub-containers
6.4.3 Definition of Indirect Measures for External Attributes
6.4.3.1 Formalize External Attributes
In the context of the present dissertation we focused on the BPMN models’ external at-tribute of correctness. The quality characteristic of correctness can be interpreted usingthe probability concept. Probability is a well understood attribute and exists a wide con-sensus about its meaning.
To formally introduce the properties regarding probability, we rely on the notion ofprobability formalized through the Kolmogorov approach (section 6.3.2.3.1). Having inmind however that it is outside of the scope of this work to delve into the probabilitytheory since here, the probability concept is used only as a auxiliary tool. A more in-depth perspective on probabilities can be found in any basic work, such as [DS02].
The definition of probability P and some of its basic properties are summarized below:
Definition 6.17. Probability – A probability P is a real valued set function defined on asample space that satisfies the following properties.
118
6. MEASUREMENT OF BPMN MODELS 6.4. Measures Derivation for BPMN Process Models
Property 6.17.1. Empty Set – Given an empty subset A of the sample space S
P(A) = 0
Property 6.17.2. Monotonicity – Given the subsets A and B of the sample space S, if A ⊆ Bthen
P(A) ≤ P(B)
Property 6.17.3. Bound – Given any subset A of the sample space S
0 ≤ P(A) ≤ 1
Property 6.17.4. Maximum – The maximum of the probability measure of the whole sam-ple space S is
P(S) = 1
Property 6.17.5. Additivity – The probability measure of the union of a finite or infinitecollection of disjoint events A1, A2, . . . is the sum of the probabilities of the individualevents
P(A1 ∪ A2 ∪ . . . ) =∞∑
i=1
P(Ai)
The link of the probability concept to the BPMN models’ external attribute correctnessis done by assuming correctness as the probability of no occurrence of well-formednessrule violation in a BPMN model. So, we need to verify the BPMN models concerningthe well-formedness rules’ violations that they may have. Hence, our sample space willbe all the BPMN models checked regarding well-formedness rules’ conformance. Theoutcome of a random experiment will be the occurrence of at least one well-formednessrule violation in a BPMN model or the absence of violations. The rules to be checkedare part of the BPMN standard and were previously defined and implemented in OCL(section 5.2). Multiple violations of the same rule are counted as a unique violation.
6.4.3.2 Define Indirect Measures
The probability of an event, in our case the violation of well-formedness rules in a BPMNmodel of the population, cannot be measured directly, but it can be estimated. The actualmeasure of a probability requires a indirect measure and must be provided a formulafor it. Depending on the measurement goal, different statistical tools can be used forproviding the estimation.
In statistics, correlation refers to a statistical relationship involving association betweenvariables. The association relationship, formally refers to generic situations in whichtwo sets of data do not satisfy a mathematical condition of probabilistic independence[MR03]. However, we must bear in mind that correlation do not mean causation.
119
6. MEASUREMENT OF BPMN MODELS 6.4. Measures Derivation for BPMN Process Models
Correlation, technically refers a specialized type of relationship between mean val-ues. There are several correlation coefficients, measuring the degree of correlation be-tween random variables. The commonest of these, in parametric tests, is the Pearsoncorrelation coefficient, denoted by ρ, which is sensitive only to linear relationships [Was04].For non-parametric tests, are used the Spearman’s rank correlation coefficient and Kendallrank coefficient [Was06]. Other correlation coefficients exists more sensitive to nonlinearrelationships [MR03].
The Pearson and Spearman correlation values can be:
• +1 in the case of a perfect positive linear relationship• −1 in the case of a perfect negative linear relationship (anti-correlation)• some value between −1 and +1 in all other cases, indicating the degree of linear
dependence between the variables.
As it approaches zero there is less of a relationship (closer to uncorrelated). The closer thecoefficient is to either −1 or +1, the stronger the correlation between the variables. If thevariables are independent, the correlation coefficient is 0. The reverse is not true becausethe correlation coefficient detects only linear relationship between two variables. In thespecial case when X and Y are jointly normal, uncorrelated means independence.
Correlation coefficient between two random variables X and Y is defined as
ρ(X,Y) =Cov(X,Y)√
Var(X)Var(Y).
The sample correlation coefficient r between two samples xi and yj is defined as
r = Sxy/√
SxxSyy
The example of correlated phenomena in this work includes the one between theBPMN models’ base measures regarding internal attributes (e.g. complexity, size) andBPMN models’ external attribute correctness. Such kind of correlations are useful be-cause they can be exploited in practice by process modelers, for predictive purposes.
We also use in this work, the Binary Logistic Regression statistical tool [CC97], which isflexible enough for modeling a number of different relationships. The BLR is a classifica-tion technique for estimating the probability of a BPMN model has to belong to a specificclass, based on the values of the independent variables that quantify the BPMN models’base measures. For calibration of the BLR model, must be collected and recorded dataon the violations of well-formedness rules. For the case of the dependent variable Y (cor-rectness of BPMN models in our case), which can take only two values, 0 and 1, and anynumber of independent variables Xi (measures of BPMN models’ internal attributes), themultivariate binary logistic regression model is defined by the equation:
π(X1,X2, . . . ,Xn) =e(β0+β1·X1+...+βn·Xn)
1 + e(β0+β1·X1+...+βn·Xn)(6.6)
120
6. MEASUREMENT OF BPMN MODELS 6.5. Conclusion
Coefficients βi are estimated through maximum likelihood estimation. The followingtwo statistics are used to describe our experimental results:• ρ – The level of significance of the binary logistic regression coefficient βi provides
the probability that βi is different from zero by chance, i.e., Xi has an impact on π.• R2 – The goodness of fit range between 0 and 1. The higher R2, the higher the effect of
the model’s independent variables, the more accurate the model. A high value forR2 is rare for logistic regression. R2 may be described as a measure of the proportionof total uncertainty that is attributed to the model fit.
6.5 Conclusion
Usually process models’ measures are derived focusing almost exclusively on their theo-retical or empirical validation. This may lead to results difficult to interpret or generalize.
In this dissertation we cover the two approaches for a set of proposed BPMN mea-sures. In this chapter, we began theoretical validation by setting up the BPMN measure-ment terminology (section 6.2), grounded on concepts and definitions anchored uponthe metrology vocabulary and the industry’s set of standards on quality, on measure-ment concepts from the discipline of Software Engineering, as well as on an the SMOontology.
We also proposed the adaption of a framework for BPMN measurement, based uponthe measure definition process GQM/MEDEA, a systematic approach from softwaremeasurement. The proposed BPMN measurement framework allowed the customizedsetting up of guidelines for design and definition of sound BPMN measures (section 6.3).
The BPMN measurement framework is instantiated in section 6.4. Based on theknowledge of the BPMN language, were explicitly defined specific measurement goals.Also a set of intuitive hypotheses were formulated in order to be validated. Next, in-ternal attributes of interest were identified and base measures were validated against aset of theoretically properties. The base measures are intend to quantify various BPMNmodels’ internal attributes (e.g. size, complexity, etc.) and are the independent variablesof a prediction model for BPMN measurement. Finally, was derived an indirect measure,quantifying the dependent variable, and related to an external attribute (correctness). Themain contribution of this chapter is the correlation relationship and the prediction modelbuilt for explaining the relationship between BPMN models’ base measures and the ex-ternal quality measure.
The results attained in this chapter will be subject of empirical validation in chapter9.
121
[This page is intentionally blank]
7Model Driven Approach for BPMN
Verification and Measurement
"Knowing reality means constructing systems of transformations that correspond,more or less adequately, to reality."
Context: The MDE paradigm advocates that relevant concepts of a domain, as well asany changes made in the domain, can be depicted as a model.Objective: Highlight the relationship that can be established between MDE and BPMN,which underpins the expected contributions of the present research work.Method: The MDE approach was instantiated for BPMN through MDA/Ecore ap-proaches.Results: BPMN well-formedness rules were implemented in the BPMN metamodel,and was collected the data needed for empirical validation.Limitations: Albeit the approach brings the BPMN to the context of MDE, using dif-ferent kinds of tools and languages, we are aware that the potential of this relationshipwas only slightly scratched.
123
7. MODEL DRIVEN APPROACH FOR BPMN VERIFICATION AND MEASUREMENT 7.1. Introduction
Conclusion: This chapter clarifies how the results attained by the two previous ones,will be operationalized for the empirical validation to be made in the two next chapters.
7.1 Introduction
Model Driven Engineering (MDE) is an approach aiming to effectively express conceptsof particular domains, as well as tackling the growth of complexity of systems develop-ment [Sch06]. According to the MDE paradigm, the relevant concepts of the domain, aswell as any changes made in a system, are amenable of being depicted in models. So, inthe MDE, every concept must be modeled.
Any change in a process must be shown in the model that represents that process.Process models can be used either to specify a process to be implemented, or to describean existing process. Explained in a simple way new processes are produced from processmodel’s specification, while descriptive process models are designed from existing pro-cesses. So, in the context of process modeling, model engineering is about: (1) producingprocess model’s specification; and (2) generating a valid process from a process model.
Process models developed at a higher level of abstraction, represent business require-ments, and they must be correct. The transformation of these processes models to a lessabstract level, for the development and implementation on a BPMS, should be made afterbeing ensured that models are correct. Therefore, before transforming models betweenlayers of abstraction we must check the correctness and consistency of our models, i.e.,ensure that BPMN models are syntactically correct and well-formed.
Process models can have their internal and external attributes quantified throughmeasures. The measures can be instantiated and quantified by applying a measurementmethod or function (section 6.3) to the constructs of the process modeling language suchas BPMN. This allows that measurement of BPMN models attributes could be achievedindependently of platform’s deployment of processes. Platform independence is anotherprinciple on which MDE is based.
Another important concept in MDE is model transformation. By transforming mod-els, the evolution of the processes is facilitated. A process model could be transformed toanother process model or to a XML dialect (e.g. XMI, XPDL) as well as to the source code(e.g. BPEL) that implements the process model functionality.
We addressed in previous chapters some contributions of this dissertation, namelythe formalization of BPMN well-formedness rules (chapter 5) and the measurement ofquality attributes of BPMN models (chapter 6). The goal of this chapter is to go furtherregarding the linkage between MDE and process modeling. We begin by introducingthe main concepts regarding MDE, as well as the close relationship between MDE andBPMN (section 7.2). In section 7.3, we instantiate the MDE approach by implementingBPMN well-formedness rules, and collect the data needed for the empirical validationthat will be performed in chapters 8 and 9.
124
7. MODEL DRIVEN APPROACH FOR BPMN VERIFICATION AND MEASUREMENT 7.2. Process Modeling in Model
Driven Engineering
7.2 Process Modeling in Model Driven Engineering
Model Driven Engineering (MDE) is a global approach that became popular, both in theresearch and industrial communities, which aims to integrate existing results and bodiesof knowledge regarding Software Engineering.
However, the MDE approach is not restricted to the development and evolution ofsoftware systems. The same concepts underpinning MDE have been adopted in otherfields of Computer Science, albeit with different perspectives and using specific terminol-ogy. Besides the already mentioned focus on designing and building software throughModelware / UML, one can find examples of the same open and integrative approach ofMDE in several other research communities/technologies, such as Grammarware / BNF,Documentware / XML, Dataware / SQL [FN05]. Likewise, one could see process modelingas part of what we can call the Processware community with, for instance, BPMN as itstechnology.
The definition of essential concepts of MDE (e.g. models, metamodels and transfor-mations), and the relations between them, has been conveyed through the concept ofmegamodel [Fav05]. A megamodel intends to define the set of entities and relations thatare necessary to model different aspects about MDE. Megamodels should have techno-logically independent representations and its concepts should be able of being validatedthrough the use of a wide range of artifacts. Among the different ways of expressingmegamodels Favre uses UML with OCL constraints [FN05]. Concepts such as decompo-sition, representation, conformance, and transformation fit in the megamodel depicted inFigure 7.1.
For each research community, the concepts that are part of the megamodel definitionare operationalized in a different way. For instance what is called a metamodel in Model-ware corresponds to a schema for Documentware and Dataware communities, a grammarin Grammarware, a viewpoint in the software architecture community [FN05], or evena directed graph in Processware. Since BPMN is also underpinned by a metamodel, wecould also address process modeling from the perspective of Modelware.
While adapting the megamodel in [FN05] to the realm of process modeling, we putat the center of the model the process (Figure 7.1). The process can be seen in severaldifferent perspectives, such as As-Is or To-Be, manual or automated, etc. The four unaryrelationships convey the typical MDE associations that can be established among modelsin the same or different levels of abstractions.
The RepresentationOf relationship has on one side the role of model regarding a specificprocess, while the other side refers to the role of the process under study (PUS). It is notexpected that a model could convey everything about a PUS. It is only expected that itcould convey the information required for the purpose it was built. Conversely, a processmodel is an abstraction of a business process. The notion of model is relative, and is notan intrinsic property of a business process [FN05].
The ConformantTo association relates models, i.e., it links a model to its metamodel. To
125
7. MODEL DRIVEN APPROACH FOR BPMN VERIFICATION AND MEASUREMENT 7.2. Process Modeling in Model
Driven Engineering
Figure 7.1: Processware Megamodel (adapted from [FN05])
understand a model one must know in what language the model is represented. Thus, isrequired a model of the language for checking whether the model is conform to the gram-mar. A model of a language is just a representation that provides answers about the lan-guage under study. In our particular case, the model of the BPMN language is the BPMNspecification document or the BPMN metamodel. On the other hand, a BPMN tool be-ing a concrete software system embodies a metamodel when it is used to check whethera given BPMN diagram is conform or not to BPMN syntactical rules. This behavior issimilar to a Java parser which also embodies a metamodel (herein called a Grammar) andis used by programmers to check whether a certain Java program is conform or not withJava syntax.
The grammar of the language is not the same as the language itself. The language isthe infinite set of all valid statements of the language. The grammar is the finite set of allgrammatical rules of the language. On the other hand a modeling language is a set whoseelements are models. Applying these notions to the realm of BPMN one can say: theBPMN grammar is the finite set of all grammatical rules of the BPMN language. TheBPMN language is the infinite set of all valid BPMN diagrams. Since the statements ofBPMN are models, BPMN is naturally a modeling language.
Introducing the concept of modeling language makes it possible to give a precisedefinition of a metamodel from the MDE perspective. A metamodel is a model of a mod-eling language. The BPMN specification document plays the role of metamodel becauseit models the BPMN modeling language. Also in the ConformantTo association emphasismust be put on the fact that the notion of model and metamodel are roles, not intrinsicproperties of processes [FN05].
To model the large diversity of systems, MDE proposes the use of Domain SpecificLanguages (DSL) [BAGaB11]. By means of DSLs, can be adopted suitable notations fordifferent system’s purposes.
Another of the key ideas of the MDE approach is that most of the transformationsbetween models can be described, and (partially) automated, thanks to transformationlanguages. The concept of transformation as a cornerstone of the MDE is depicted bythe unary association IsTransformedIn in Figure 7.1. In the context of this work, source
126
7. MODEL DRIVEN APPROACH FOR BPMN VERIFICATION AND MEASUREMENT 7.2. Process Modeling in Model
Driven Engineering
process models can be transformed into target models, either graphical or textual (e.g.XML dialect serialization) that implement another perspective or functionality [PRP08] ofa business process. The unary association DecomposedIn embodies the ability of a processto be composed of elements with less complexity, or be the composition of elementaryconstructs.
Some of the most widely-known MDE instantiations are the OMG’s Model DrivenArchitecture (MDA)1 initiative and the Eclipse2 set of modeling and programming lan-guages.
7.2.1 BPMN in the context of MDA
The MDA standard was seminal in the realization of MDE. MDA is based on the use ofabstract representations called models. Raising the level of abstraction from code produc-tion to models is a core idea of MDA. The notion of platform played also an importantrole in MDA through the building of models independent from platforms. Another im-portant characteristic of MDA is the use of metamodels and metamodeling techniques asa mean both to describe languages and to structure artifacts.
The MDA faced challenges common to the industrial standards adopted by largecommunities. The incremental definition, reflecting the consensus achieved among con-tributors, affected MDA overall quality when compared with well grounded techniquessuch as formal specifications [Fav04]. One of the issues pointed out to MDA standardis that there is no clear separation between essential concepts and the technologies thatimplement these concepts [Fav05].
Explicit metamodels are a prerequisite in MDA, given its emphasis on automation.The model concept has a more restrictive definition in the realm of MDA, than the onegiven in MDE. Here, the concept of model is strongly connected to the notion of meta-model. In MDA a metamodel is a model that defines the language for expressing othermodels [OMG03b]. According to the MDA standard all metamodels must be written inthe Meta Object Facility (MOF) language to be MDA compliant.
Therefore, MOF is a key instrument for MDA. MOF ensures that all compliant meta-models share a common set of core assumptions and definitions. MDA models are relatedsince they all are mapped upon the constructs of the MOF abstract meta-metamodel. Ev-ery model or metamodel used in MDA, is directly or indirectly defined in terms of MOFconstructs, being therefore considered as MOF compliant (see Figure 7.3). This ensuresthat the concepts of all models used in MDA are able to be transformed in their equiva-lents in every other MOF-compliant model. MOF borrows a tiny subset of UML’s largeset of constructs for the purpose of modeling a language. The OMG has produced MOFmodels for a number of languages. Examples of MOF-compliant metamodels, are the
1http://www.omg.org/mda/2http://www.eclipse.org/
127
7. MODEL DRIVEN APPROACH FOR BPMN VERIFICATION AND MEASUREMENT 7.2. Process Modeling in Model
Driven Engineering
OMG’s Common Warehouse Common Warehouse Metamodel (CWM) [OMG03a], Se-mantics of Business Vocabulary and business Rules (SBVR) [OMG12a], Ontology Defini-tion Metamodel (ODM) [OMG09], and also BPMN [OMG11]. Another MOF-compliantconcept is the UML Profile. Profiles are extensions supported by UML, used to describevarious functional uses of UML. Being extensions of UML, profiles are also considered asMOF metamodels.
Since, the BPMN metamodel is defined in the MOF language, BPMN language isMDA compliant. Similarly, all the diagrams included in BPMN (e.g. conversation, chore-ography or orchestration diagrams) are amenable to be transformed in other MOF-compliantlanguages.
Even languages that are not based on UML can still be MDA languages if they satisfythe MDA requirement, which is to have a MOF model of the language itself. MOF allowsmanaging models in an integrated manner, even when the models are expressed in adisparate language. Therefore, MOF presents an approach to metadata integration moreeffective than the efforts based on only one modeling language [Fra03].
MDA’s perspective regarding processes is based on two principles: (1) the emphasison models; and (2) the transformation of process models with metamodels’ mappings. Byfollowing these principles what is attained are processes’ representations where the do-main logic specification is separated from the platform specific details in which processesare implemented.
The emphasis on models is achieved by layering models processes’ developed ac-cording to three levels of abstraction: the Computation Independent Model (CIM), thePlatform Independent Model (PIM), and the Platform Specific Model (PSM) (see Figure7.2). Mappings among models are necessary to move between levels of abstraction andwithin the same level of abstraction. This can be operationalized through upward ordownward navigations between metamodels. For instance, if we want to transform aPSM model into another PSM model, we can move using mappings from the source PSMmodel to a common metamodel and then to the target PSM model.
MDA primary focus is on the development and maintenance of software artifacts.MDA also describes how models are used in the software development process, from thehighest level development model in CIM to the platform oriented PSM [Har04]. CIMmodels (e.g. Use Cases) are used to describe the problem. PSM models (e.g. Class orSequence Diagram incorporating specifications of implementation environments such asJ2EE or .NET) describe the solution.
We can draw a parallel between the life cycles of software development and businessprocesses to figure out how the latter can be driven by MDA. Business managers andprocess analysts demand more business oriented artifacts than the ones used by softwaredevelopers. Therefore, this perspective should be provided through a more businessfamiliar view of process models, built using a process modeling language such as theBPMN. With BPMN the actual business processes realized in the organization, can beelicited through diagrams with details according to the different levels of abstraction. In
128
7. MODEL DRIVEN APPROACH FOR BPMN VERIFICATION AND MEASUREMENT 7.2. Process Modeling in Model
Driven Engineering
Figure 7.2: BPMN diagrams in MDA levels of abstraction
fact, BPMN has some advantages for business processes modeling elicitation accordingto the MDA, when compared to the UML, namely the usage of the same notation forCIM, PIM, or PSM modeling. The difference among the BPMN models at the differentlevels of abstraction, resides on the level of detail adjusted to the intended usage.
CIM models are BPMN diagrams developed by domain analysts mainly for docu-mentation purposes and to provide a broad knowledge of the business processes’ do-main. CIM just describes concepts related to the particular domain, with no referenceto the particular problem to be solved in that domain. BPMN public processes are usedat CIM level since they are usually modeled to show only the relevant flow elements ofthe collaboration made with external entities. Therefore, since the internal details of pro-cesses are omitted from the model they are used essentially for documentation purposes[BPM11, page 147].
On the other hand, through PIM models, architects and designers incorporate detailsof business processes beyond the happy path, such as exceptions flows, events, and com-pensations, but in a technology independent manner. PIM models are more precise andcan be used as specification of a business process. What they say about the business pro-cess is expected to be true, although they describe only some aspects of it. The definitionof PIMs is made once and does not need to be made for all technologies, which allowsdesigns portability and interoperability of processes to be defined at model level. Con-versely to CIM, where process models were merely a process design artifact, intended tohelp human process managers design and understand processes, at PIM level the processmodels must be sufficiently formal and detailed for being used as a runtime artifact thatcould fuel up applications such as simulators and process tuners.
129
7. MODEL DRIVEN APPROACH FOR BPMN VERIFICATION AND MEASUREMENT 7.2. Process Modeling in Model
Driven Engineering
Finally, PSM models embody the details required by each particular deployment plat-form and are specific of each BPMS. Thus, PSM models must be rigorously formalizedand must contain enough information to be directly interpreted by BPMS engines or al-low the derivation of executable processes in a language such as BPEL or web services.Mappings between specific PSM models and other programming languages depend onthe availability of profiles that rigorously specify how a given process model should betransformed [OMG13, MB02]. Various types of mappings between the different levels ofabstraction are possible (see Figure 7.2), namely via XMI or MOF.
In the context of collaborative business process management, MDA standards playthe role of integration and allow different organizations to cooperate from a businessprocess point of view [PRP08]. MDA is also relevant for business processes consider-ing the deployment platform. Business processes are deployed and automated throughBPMS, on top of software components providing process model compilers, interpreters,debuggers, and so forth. Changes in business processes have impacts on software com-ponents. So, business processes’ effectiveness and efficiency heavily rely on the softwaresystem’s suitability to keep in sync with domain requirements, which is, ultimately, theaim of MDA.
Tracing models from CIM to PSM process models became more simple than usingUML disparate sorts of diagrams. MDA allows transformation, via MOF, of BPMN dia-grams to their equivalent diagrams in other languages or notations.
However, though understandable and intuitive, the notions of platform, PIM andPSM are not clearly defined in the MDA standard, and therefore they are subject to somecontroversy. Albeit there is a continuum between PIMs and PSMs, the distinction be-tween these models is not clear-cut [Fav05]. As it happens in the case of Software En-gineering, process modeling in the context of MDA also requires mathematically-soundmethods that enable stepwise refinement from process models to executable processes.Moreover, the validity of the transformation should be demonstrated at each step.
The transformation among models, written for instance in a MOF-compliant lan-guage (e.g. Query/View/Transformation (QVT)), is also intrinsic to the MDA paradigm.Model’s transformations are performed based on metamodels’ mapping and the generictransformation architecture depicted in Figure 7.3. For instance, suppose is required atransformation from an UML activity diagram (source model) to a BPMN process di-agram (target model). Starting first from the UML activity diagram, a match betweenequivalent concepts should be found at MOF compliant metamodels, between UML(source metamodel) and BPMN (target metamodel). Lastly, after the identification of themappings among equivalent constructs in both metamodels, i.e. for the initial conceptsof UML activity diagram matching the related BPMN graphical constructs, one shouldbe able to set up the transformation.
The BPMN standard has a metamodel that contains the syntactical definition of thelanguage’s constructs, as well as the logical relationships among these constructs (Figure3.4). The graphical notation (Figures 3.2 and 3.3), supported by BPMN tools, defines the
130
7. MODEL DRIVEN APPROACH FOR BPMN VERIFICATION AND MEASUREMENT 7.2. Process Modeling in Model
Driven Engineering
Figure 7.3: MDA transformation architecture
visual representation of process models, making them more suitable for human readingand understanding [Fra03]. The output of modeling can be serializable and saved inXML like format (XMI/XPDL). The exchanging of process models and their diagramscan be done through the standardized exchange formats, which enables the transferenceof models created in specific BPMN tools to other BPMN tools or software languages[Lon04].
Since XPDL is based on BPMN, its metamodel is close to the BPMN’s meta-model.Some elements are added in XPDL to facilitate implementation [Gao06]. In XPDL a pack-age corresponds to the BPMN Business Process Diagram, and consists of a set of processes’definitions. A process definition has several activities. Activities are related to each otherthrough transition information. The participant declaration provides the resources to per-form activities, while the application declaration describes IT applications invoked in thesystem. The relevant data field defines the data that is created and used within each processinstance during process execution.
The graphical notation, as well as the serial representation, are tightly related to theBPMN metamodel, and must conform with it. The BPMN standard as a process modelinglanguage is the key for attaining business process modeling’s goals, as well as achievingprocess models’ platform independence. Platform independence is exactly one of theprinciples advocated by the MDA framework [PRP08]. The BPMN standard follows theMDA approach, since it is based on a MOF compliant metamodel which allows transfor-mations’ definition for process models defined in the language’s notation.
The quality of process models developed at each layer of the MDA pyramid, mustconform to the requirements of the adjacent layer. Therefore, the transformation of pro-cess models from PIM to the PSM or from PSM to specific BPMS, should ensure theprocess models’ compliance with predefined internal and external attributes. The toolswe presented in chapters 5 and 6 intended to contribute for attaining these objectives.
131
7. MODEL DRIVEN APPROACH FOR BPMN VERIFICATION AND MEASUREMENT 7.2. Process Modeling in Model
Driven Engineering
7.2.2 Model-based Testing of BPMN Models
In chapter 5 we formalized BPMN well-formedness rules, to ensure the soundness of ourproposal. The previous section allows us to position BPMN process modeling artifacts inMDA. In this section we still follow a model-driven approach by characterizing a model-based testing for BPMN. Here, we intend to define the principles to set up the tests thatwill be conducted in section 7.3. This is the basis to empirically validate the effectivenessof our proposal in chapters 8 and 9.
Before presenting the BPMN model-based testing framework for verifying the BPMNwell-formedness rules (see Figure 7.4), we introduce some terminology concerning test-ing in the context of BPMN process modeling, as well as some aspects regarding datacollection of faults in samples of BPMN models. Most of the concepts used were adaptedfrom the realm of software testing [BDG+08].
One of the purposes of process modeling testing in the context of this dissertation is todetect BPMN models’ failures so that data collected can be statistically processed. Testingis seen here as a mean for comparing the compliance of a BPMN model with specific well-formedness rules prescribed by an oracle – the BPMN specification standard [BPM11].
Process models’ faults occur, for instance, when a process modeler, using BPMN con-structs, violate well-formedness rules. This results in a defect in the process model rep-resentation. If the process model’s defect is present when an actual business process isexecuted by a BPMS, wrong results could be generated, causing failures or inconsisten-cies in the business process.
In the context of BPMN process modeling, a test suite is a collection of test casesthat are intended to be used to test process models and show whether they comply witha specified set of well-formedness rules. The test suite defines goals for the collection oftest cases and embodies information regarding the testing environment configuration to beused during tests. The testing environment is a setup of software applications on whichthe process tester is going to perform the testing of process models.
In this work BPMN models are submitted to what can be called as static testing. Statictesting is a kind of testing, by which process models do not actually run business pro-cesses. In static testing process models are analyzed by an interpreter that asserts theprocess models’ correctness. Static testing can be made before the process model is in useor even before the model is fully designed, in order to verify particular parts of the model(e.g. subprocesses). Bugs that are eventually discovered at this preliminary stage of busi-ness process design are less expensive to fix than the ones discovered at later stages (e.g.deployment).
The tests conducted on BPMN models, since they are model-based testing, can alsobe considered black-box testing. The process modelers (process analysts or process imple-menters) test process models by examining their compliance with BPMN specificationand disregarding any knowledge about how well-formedness rules were actually imple-mented. This testing method can be applied to different levels of detail of process models:
132
7. MODEL DRIVEN APPROACH FOR BPMN VERIFICATION AND MEASUREMENT 7.2. Process Modeling in Model
Driven Engineering
subprocesses, the overall business process, or collaboration among processes.The BPMN model-based testing framework presented in Figure 7.4 can be seen as
another incarnation of model-driven engineering [UL06]. This testing approach will beinstantiated in the next section through the design and execution of a set of test casesagainst an enhanced version of the BPMN metamodel – the System Under Test (SUT).
Figure 7.4: A framework for BPMN model-based testing
Model-based testing is used here for comparing the SUT with specifications. TheBPMN standard document [BPM11] is the specification of the SUT. The standard docu-ment provides information to build the necessary test cases for detection of faults in theimplementation of BPMN well-formedness rules. There is traceability between each testcase and the correspondent rule specification.
In BPMN model-based testing framework, we distinguish between abstract test suites,which are collections of test cases derived from the BPMN metamodel – the system undertest – and executable test suites, which are derived from abstract test suites by providinglower-level details needed for a test case to be executed by an interpreter of BPMN mod-els’ instances. The abstract test suite cannot be directly used because it remains at a highlevel of abstraction and lacks details about the execution environment.
The abstract test suite, which is composed by BPMN models describing the SUT re-quirements, depicts only a portion of the SUT’s desired behavior. Each abstract test casebehaves as a functional test on the same level of abstraction of a regular BPMN model.Since the abstract test suite is also at a different abstraction level of the SUT, the executabletest suite needs to be derived from the abstract test suite. The transformation from theabstract test cases to concrete test cases provides the essential information for the exe-cution of the latter in the testing environment. The executable test suite can then runagainst a representation of the system under test for checking whether well-formednessrules stated in the specification are well implemented.
133
7. MODEL DRIVEN APPROACH FOR BPMN VERIFICATION AND MEASUREMENT 7.3. Instantiation for BPMN
Verification and Measurement
Setting up the framework for testing the SUT establishes also the foundations for thetestbed environment to be used on the verification soundness of hypotheses regardingBPMN models’ quality characteristics made in chapter 6. This will be done through theempirical validation of the hypotheses using a sample of BPMN models as test scenario.
By creating the mentioned model-based testing framework, we intended to avoid aninformal testing approach based on ad hoc tests. Instead, our aim was to automate testsas much as possible, however without giving up the human intervention to monitor andanalyze produced errors. Regression tests through the process were also essential to beconsidered. The aim was to ensure that previous well-formedness rules implementeddid not suffer collateral effects from the updates in the metamodel due to new ones im-plemented and tested.
7.3 Instantiation for BPMN Verification and Measurement
In the present section, we use BPMN at different levels of abstraction of MDE, from themetamodel layer (M2) to the instance layer (M0). We address BPMN models’ specifica-tion at each abstraction level, as well as the transformation between layers.
The work described in this section contributes to the verification of well-formednessrules of BPMN language formulated in chapter 5, as well as for collecting data for em-pirical validation of BPMN models conformance with those rules to be used in chapter 8.This section also contributes for the validation of BPMN measures proposed in chapter 6and empirically validated in chapter 9. The overall process, which will be detailed in thefollowing sections, consists in several steps.
The BPMN metamodel was firstly imported and converted from a XMI representationto a script in the concrete syntax of a tool checker. The BPMN models produced by BPMNdesign tool were transformed also into the concrete syntax of the tool checker. While thescript was interpreted by the tool, instances of BPMN models were validated. Hence, theapproach transforms BPMN models to the concrete syntax of a BPMN model checker forBPMN well-formedness rules verification and BPMN models measurement. These stepsoccur at early stages of the design process with BPMN models expressed at PIM level.A BPMN model that positively passes the checking process means that it is accuratelyexpressed in the BPMN language.
Since the models are at a high level of abstraction, this approach contributes alsofor bridging the gap within the BPMN community between domain analysts, who workwith processes at a domain level (CIM), and process implementers, who analyze the sameprocess at a technical and implementation level (PIM).
134
7. MODEL DRIVEN APPROACH FOR BPMN VERIFICATION AND MEASUREMENT 7.3. Instantiation for BPMN
Verification and Measurement
7.3.1 BPMN Models’ Verification and Measurement
The OMG BPMN metamodel describes the abstract syntax of the BPMN language bymeans of meta-classes, meta-associations and cardinality constraints. We started by check-ing BPMN model syntax by instantiating the BPMN metamodel in the USE validationenvironment [GBR07]. This tool allows checking whether a set of objects and their linksmatch the corresponding model structural constraints, namely regarding cardinality andtype conformance, as well as OCL invariants concerning BPMN well-formedness rules.
To operationalize the aforementioned objective, we developed the M2/M1 transfor-mations depicted in the process model of Figure 7.7:• from the BPMN metamodel (available in XMI format in the OMG site) to the USE
abstract syntax. This transformation consists in importing the XMI file into a CASEtool (Enterprise Architect) repository and then, using the Java API of the CASE tool,generating a file with the BPMN metamodel in the human-readable textual formatof USE abstract syntax.The same Java API of the CASE tool was used for building the EA2USE transfor-mation (M1/M0) (Figure 7.5). The EA2USE is aimed to convert to USE concretesyntax: (1) BPMN models snippets for testing well-formedness rules, as well as(2) the BPMN models produced by surrogates of process modelers for the quasi-experiment described in section 8.5.The USE concrete syntax is a set of commands to instantiate a process model in theUSE environment, ensuring its conformance to a previously loaded metamodel.• converting the XMI file into an Ecore BPMN metamodel, for creating an Ecore USE
metamodel, and Ecore XPDL metamodel. The Ecore metamodels were expressedusing the semantics of the Ecore metametamodel.Using the two Ecore metamodels was built an ATL [Ecl11] transformation (XPDL2USEin Figure 7.6). The XPDL2USE transformation (M1/M0) aimed to convert BPMNprocess models collected from open repositories, conform the XPDL metamodel,into the equivalent USE concrete syntax . The details of those process models, se-rialized in XPDL files and transformed for the USE environment are described insection 8.4.
These transformations had to match USE language conventions, thus requiring someminor changes in meta-association names such as appending as suffix an underscore plusthe alphabetic character a to identifiers which are reserved keywords in USE (e.g., opera-tions, from), or an underscore plus an alphabetic character (a, b, or c) to the target/sourceto the role identifiers of associations between the same meta-classes.
After having been accomplished the BPMN metamodel transformation to the USEabstract syntax, the file with the transformed metamodel could be loaded by the USEenvironment. Figure 7.9 shows the USE tool loaded with the 151 meta-classes and 200meta-associations (see the Log window) of BPMN. The Class diagram window shows acluttered snapshot of the corresponding class diagram.
135
7. MODEL DRIVEN APPROACH FOR BPMN VERIFICATION AND MEASUREMENT 7.3. Instantiation for BPMN
Verification and Measurement
Figure 7.5: The EA2USE transformation tool
Figure 7.6: The XPDL2USE transformation
136
7. MODEL DRIVEN APPROACH FOR BPMN VERIFICATION AND MEASUREMENT 7.3. Instantiation for BPMN
Verification and Measurement
Figu
re7.
7:Bu
ildin
gBP
MN
synt
axva
lidat
orth
roug
h:La
ne1
–th
etr
ansf
orm
atio
nof
the
BPM
Nm
etam
odel
into
the
USE
abst
ract
synt
axan
dth
eco
nstr
ucti
onof
EA2U
SEtr
ansf
orm
atio
n;an
dLa
ne2
–th
eco
nstr
ucti
onof
XPD
L2U
SEtr
ansf
orm
atio
n
137
7. MODEL DRIVEN APPROACH FOR BPMN VERIFICATION AND MEASUREMENT 7.3. Instantiation for BPMN
Verification and Measurement
Figu
re7.
8:Bu
ildin
gan
dve
rify
ing
BPM
Nm
odel
ssn
ippe
tsfo
rBP
MN
wel
l-fo
rmed
ness
rule
sde
riva
tion
138
7. MODEL DRIVEN APPROACH FOR BPMN VERIFICATION AND MEASUREMENT 7.3. Instantiation for BPMN
Verification and Measurement
Figure 7.9: The USE environment loaded with BPMN metamodel and the BPMN diagrampresented in Figure 7.10
Figure 7.10: A BPMN simple diagram of a transactional sub-process
Subsequently, BPMN models snippets were built with the mentioned CASE tool. Thedepicted elements’ definitions were exported through EA2USE transformation used toget the BPMN models instances definitions equivalent in the USE concrete syntax (seelanes of participants Enterprise Architect and EA2USE in Figure 7.8).
We were then able to instantiate the BPMN metamodel with BPMN models instances,as it can be ascertained in the Object diagram window in Figure 7.9 where one can see acluttered snapshot of the meta-object diagram corresponding to the BPMN model extractin Figure 7.10. The Command list window shows the commands issued to create the in-stances of meta-classes and meta-associations, as well as to set their state. Object countand Link count windows display the number of instances of meta-objects and meta-links
139
7. MODEL DRIVEN APPROACH FOR BPMN VERIFICATION AND MEASUREMENT 7.3. Instantiation for BPMN
Verification and Measurement
created, by type.By this time, were caught syntactical errors by the USE tool. Examples of such type
of errors include among others, typeless instances and connections among elements notallowed in the metamodel, such as a DataInputAssociation linking two instances of Task,an instance of MessageFlow linking an instance of Gateway to an instance of Task.
The next step was to instantiate the framework for BPMN model-based testing (see Fig-ure 7.4 in section 7.2.2). Building this environment allow us to validate BPMN processmodels through a BPMN metamodel enriched by the addition of well-formedness rulesas OCL invariants, corresponding to the informally conveyed rules throughout the OMGspecification, complemented with best practices from the field.
In Figure 7.8 we depicted the overall process model with the activities taking placeto enhance the BPMN with the mentioned rules. The JUSE-JUnit lane refers the role of aJava facade3 for the USE tool used for rules testing and debugging. After each rule wascodified, added to the BPMN metamodel and syntactically validated (lanes Researcherand USE), a BPMN model snippet (test case) was build (lanes Enterprise Architect andEA2USE) for checking the correctness of the rule.
We elicited 133 invariants (Appendix E) and implemented 610 operations, resultingin a total of 743 OCL expressions classified as follows:• Flow Control Well-formedness Rules: rules related with the interaction among model-
ing elements;• Data Flow Well-formedness Rules: rules related with sharing of data by activities;• Best-Practices Recommendations: optional rules related with advised usage of BPMN
elements in diagrams.The Class invariants window in Figure 7.9 shows the results of the model check per-
formed upon the BPMN model of Figure 7.10. One can also see that was broken at leastone well-formedness rule (denoted by the boolean value false). By querying the brokenrule we can understand its semantics: a throwing compensate event is not allowed in atransitional sub-process.
7.3.2 Data Collection for Empirical Validation
An empirical study was conducted to determine the BPMN models’ external attribute ofcorrectness (conformance of BPMN models with the BPMN specification), as well as themeasurement of BPMN models’ internal attributes (depicted in Figure 6.2). The sampleused (see details in section 8.4), was based upon available BPMN models stored in publicrepositories managed by two BPMN tool providers:• BizAgi4,5 - a Business Process Management solution provider, positioned in the
2010 Gartner’s BPMS Magic Quadrant [HCKP09], which made available on-line 19customizable templates of business process models;
3http://code.google.com/p/j-use/4 http://www.bizagi.com/5BizAgi is one of the copyright holders of the BPMN 2.0 standard.
140
7. MODEL DRIVEN APPROACH FOR BPMN VERIFICATION AND MEASUREMENT 7.3. Instantiation for BPMN
Verification and Measurement
• Trisotech6 - a provider of consulting services and BPM solutions, which runs anon-line resource repository, the Business Process Incubator, with almost 50 BPMNbusiness process models collected from several sources.
Figure 7.11 depicts the activities of data collection that took place for the empiricalstudy analysis. Those activities are briefly described below:• Each of the business process models was downloaded from the respective site: (1)
BizAgi models were in a proprietary format used by the tool (BizAgi Process Mod-eler v.2.3) of the repository owner; (2) business process models from Business Pro-cess Incubator were in Visio format. These models were converted to BizAgi formatsince BizAgi Process Modeler can import Visio files and save them in BizAgi ownformat;• After having all the files in BizAgi format, it was possible to convert them, using
a functionality available in the BizAgi tool, to XPDL 2.2 format7, a standard fromWfMC, which allows the serialization of business process models and the exchangeof process definitions;• Having business process models samples serialized into XPDL format, in order to
make their verification for possible standard or best-practices violations, we con-verted the XPDL concrete syntax to USE concrete syntax, using the transformationtool XPDL2USE (Figure 7.6).• The business process models now expressed in the USE concrete syntax are verified
against syntactic and well-formedness rules present in the BPMN metamodel readinto the USE tool. Any syntactic or well-formedness violations were output to afile.• Measure observations of the BPMN process instances were collected from the out-
put to a file produced by the USE tool checker, which executed OCL operations forcomputing the measure’s instances;• At the end of the BPMN models’ verification, the statistics were consolidated into a
file that was imported by the IBM-SPSS statistical tool for data analysis.A second empirical study (detailed in section 8.5) was conducted through a quasi-
experiment to evaluate the effectiveness of using automatic BPMN rule checking for as-sessing the correctness of BPMN process models. The process model of this empiricalstudy follows closely the one depicted in Figure 7.11, with some minor differences suchas: (1) the BPMN models were built by process surrogates using the CASE tool; (2) thetransformation tool used was the EA2USE; and (3) measures observations were not col-lected in this case.
7. MODEL DRIVEN APPROACH FOR BPMN VERIFICATION AND MEASUREMENT 7.3. Instantiation for BPMN
Verification and Measurement
Figu
re7.
11:D
ata
Col
lect
ion
ofBP
MN
proc
ess
mod
els
for
empi
rica
lval
idat
ion
142
7. MODEL DRIVEN APPROACH FOR BPMN VERIFICATION AND MEASUREMENT 7.4. Conclusion
7.4 Conclusion
Within the MDE paradigm, every concept must be explicitly modeled. BPMN modelsrepresent business requirements, and they must be correctly expressed. The transforma-tion of BPMN models for development and implementation on a BPMS, should be madeafter ensuring that BPMN models are correct. BPMN models can have its attributes quan-tified through measures. This allows the measurement of process models independentlyof business processes’ deployment platform. Platform independence is another principleon which MDE is based. Another important concept in MDE is model transformations.A process model can be transformed to another process model or to a XML dialect as wellas to an executable process.
This chapter introduced the main concepts regarding MDE, brings BPMN to the con-text of MDE, and presents a framework for BPMN model-based testing (section 7.2). TheMDE approach was instantiated in section 7.3 by BPMN well-formedness rules imple-mentation, and data collection for the empirical validation to be presented in chapters 8and 9.
143
[This page is intentionally blank]
8Empirical Studies on BPMN
Verification
"In so far as such a theory is empirically correct it will also tell us what empirical factsit should be possible to observe in a given set of circumstances."
Context: The current version of the BPMN standard has rules that are difficult to fol-low and enforce, which makes difficult to build BPMN models without the appropriatesupport of verification tools.Objective: To empirically validate BPMN well-formedness rules, previously formal-ized, in order to assess whether these rules could have a significant impact upon thequality of BPMN models.Method: Empirical studies are conducted using a specific framework, conform to thescientific method.Results: Our approach unveiled a more effective BPMN models’ verification than check-ing models through regular BPMN modelers or actual BPMN tools.
145
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.1. Introduction
Limitations: The samples used in the experiments were collected from examples hostedin open repositories, as well as BPMN models built by students, as surrogates of profes-sional BPMN process modelers.Conclusion: More research is needed in order to corroborate the outcomes of the exper-iments. So, the replication of the studies in industry should be the next step to ensurethe generalization of results.
8.1 Introduction
One would expect of BPMN modelers regardless of their technical background, eitherprocess analysts or process implementers (section 2.5.2.1), could be able to produce cor-rect process models using currently available tools.
The BPMN 2.0 standard is based in a metamodel with large number of meta-classes(section 3.3) and several different kinds of graphical elements and stereotypes (Figures 3.2and 3.3), constrained by more than a hundred well-formedness rules. Harnessing all thepotential and expressiveness of the BPMN standard made available by such a plethoraof elements, and building correct BPMN models, does not seem to be an easy task.
On one hand, process analysts use the language for purposes such as a documentationor process improvement and have to deal with the subtleties of exception and parallelpaths, as well as the multiple kinds of interrupting and non-interrupting events. On theother hand, process implementers, besides having to deal with previous challenges, needalso to face the complexity of the language when detailing and tuning process models forenactment upon a BPMS engine. Thus, both roles have to ensure that resulting processmodels are syntactically correct and well-formed. One would expect that a means to en-force the verification of compliance with well-formedness rules, such as the formalizationmade in chapter 5, would have a great impact upon the final quality of process models.
In section 8.2 we discuss some of the research methods available for a grounded re-search, as well as the motivation for our choice upon the scientific method. Next, wedescribe two empirical studies that validate the work done regarding BPMN models’verification (sections 8.4 and 8.5). The empirical studies’ structure follows closely theguidelines for reporting experiments proposed in [JP05]. The inception of the empiricalstudies is made from a common basis (section 8.3). Thenceforth, each step of the empiri-cal studies is reported separately in its own section, for the sake of clarity and focus. Theexperimental design is described (sections 8.4.1 and 8.5.1), namely regarding the goalsand hypotheses tested. After the studies’ execution (sections 8.4.2 and 8.5.2) the resultsare presented (sections 8.4.3 and 8.5.3) and interpreted (sections 8.4.4 and 8.5.4). Finally,conclusions are drawn in section 8.6.
146
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.2. Choosing the Research Method
8.2 Choosing the Research Method
The process of making Science requires activities such as collecting and analyzing data,interpreting the results, and eventually returning back the findings to the community.This process should be replicable by other researchers acting under the same conditions.
The selection of the research method is crucial for drawing conclusions about a phe-nomenon. The research method determines the resources needed for the study, as wellas the perspective of analysis of the factors and causes that shape the phenomenon.
To conduct science ethically, the researcher has to reveal the followed research ap-proach, and consequently the aim, theory, and method that drives the research from itsinception to the results’ communication.
Several research approaches have been used in the realm of Computer Science, Soft-ware Engineering and Information Systems. We will characterize three of the most re-current ones (action research, design method, and scientific method) in next section, to helpsetting the context for justifying the selected research approach.
8.2.1 Presenting Research Methods
8.2.1.1 Action research
Action research is an iterative research process joining researchers and practitioners, actingtogether on a particular cycle of activities, including problem diagnosis, action interven-tion, and reflective learning [ALMN99]. Action research combines theory and practicethrough change and reflection in an immediate problematic situation within a mutuallyacceptable framework. When following this approach, researchers are encouraged to ex-periment through intervention and to reflect on the effects of their intervention and theimplication of their theories [SLT09].
Action research methods can be used in application domains where research takesplace in actual organizations. Disciplines from the realm of applied science (e.g. BusinessProcesses Management, Software Engineering and Systems Science) use this approachsince it can address complex real-life problems for which practitioners seek immediatesolutions.
In action research, the researcher seeks a theory in collaboration with practitionersin real situations, gains feedback from the experience, modifies the theory as a resultof feedback attained, and tries it again. Each iteration of the action research process isexpected to corroborate (or refute) the theory.
In our research work the aim is to enforce the quality of BPMN models. So, we for-malized in chapter 5, a set of BPMN well-formedness rules. As consequence were de-fined prescriptive rules in the BPMN standard. We want now, a posteriori, using well-formedness model checking and without intervening in the modeling process, to assess thequality of BPMN models built by process modelers. To attain this goal, the action researchstudy is not the more adequate research method. Action research only is justifiable if we
147
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.2. Choosing the Research Method
wanted to assess how process modelers are using BPMN language for process modelingin organizations, being direct part in the study, which is not the case.
8.2.1.2 Design method
The Design method approach, according to Herbert Simon [Sim96], is concerned with howthings ought to be. Design demands skills and competences besides the ones derived fromscientific knowledge. Other relevant skills come from craft knowledge, such as appren-ticeship, experience, as well as trial and error. This mix of valences is necessary to designsolutions compliant with the organizational contexts and accepted as effective by prac-titioners [CNW81]. So, design is viewed more as a technological, rather than a scientificactivity.
In design methods, tacit knowledge seems to be the foundation for producing objectsfitting appropriately the users’ context. Historically, design has arisen from craft and em-bodies the know-how. The artifacts do not exist exclusively grounded upon theory. Notrarely an artifact’s invention precedes the underlying theory. Applying blindly proce-dures can hinder creativity that enables reaching the subtlety of a well-designed artifact.
It is more suitable to regard design as a technology, since design involves the applica-tion of types of knowledge other than the one that could be formally expressed. Designmethod’s activities target the invention of things of value that do not yet exist, whichmakes it constructive by nature.
The kind of activities which take place when doing the design method, are clearlyvery dissimilar from the activities that we must follow in our research work, more in linewith the scientific method, as we will see in the next section.
8.2.1.3 Scientific Method
For the Oxford English Dictionary1 the scientific method consists . . . in systematic observa-tion, measurement, and experiment, and the formulation, testing, and modification of hypotheses.The scientific method has evolved to ensure that researchers make discoveries groundedupon logic and reason [Jar01]. It relies upon a standard protocol followed by researchers,when conducting scientific research, for benchmarking and measurement of the validityof the results obtained. By following the set of procedures, researchers seek and docu-ment relationships among variables (e.g. merely associations, cause and effect relations).This means that, they raise the evidence (at some level of confidence) about the nature ofrelationships, among independent and dependent variables. The procedures of the sci-entific method are targeted at finding out the nature of the phenomena of study. Findingsare fed back into the theory that tries to explain the world.
Due to the difficulty on attaining reliability and validity of experiments carried out, dif-ferent disciplines follow the scientific method with different adjustments. Nevertheless,they all perform generically the activities described below [Shu09].
1http://oxforddictionaries.com/
148
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.2. Choosing the Research Method
1. Research Problem Identification – after the formulation of the general question aboutresearch domain, it proceeds with the observations and measurements regardingthe phenomena, collecting and organizing empirical facts. General questions aremade, narrowing down the focus to a particular aspect which could facilitate theformulation of realistic hypotheses concerning the phenomena.
2. Hypotheses Formulation – an activity supported by the inductive reasoning which al-lows researchers to come up with general principles and the elicitation of hypothe-ses with properties of testability and falsifiability [Pop35].
3. Experimental Design – consists in making predictions, using deductive reasoning, abouthow things should behave, as well as in developing a procedure for reliable ex-periments. The experiments should be designed with statistical tests in mind, bymaking sure that they have appropriate controls and a sufficiently large sample toprovide statistically valid results. The design of experiments establishes the stepsthat will test and evaluate the hypotheses, manipulating one or more variables togenerate analyzable data. Furthermore, data and observations must be recordedin order that the experiments could be replicated. The exact replication and veri-fication of the experiment by independent researchers ensures the reliability of theresults.
4. Hypotheses Testing – is the use of statistics to determine the probability that a givenhypothesis is true. The usual process of hypothesis testing consists of: (1) formulat-ing the null hypothesis and the alternative hypothesis; (2) identification a statistictest to assess the truth of the null hypothesis; (3) computing the significance of thestatistic test; (4) deciding for the acceptance or rejection of the null hypothesis.
5. Interpretation and Conclusions – the researchers provide interpretation of the resultsand, through deductive reasoning, perform conclusions about their experiments. Inlight of the information gathered during the experiments, evidence is obtainedabout the truth or falseness of the original hypothesis. If truth, the hypothesis isnot rejected and remains as a possible explanation of phenomena allowing the gen-eralization of the findings. If not, researchers reject the hypothesis and try to comeup with alternative, i. e., refined hypotheses, for the phenomena’s explanations.Whether a researcher’s initial hypothesis is right or wrong isn’t as important aswhether he sets up well-designed, repeatable experiments that provide necessaryinformation to contribute to the advance of scientific knowledge.
6. Results Communication – consists in sharing the results with the peers and scientificcommunity in general through the appropriate forums and journals allowing otherresearchers the replication of the findings and future development of the research.A rule of thumb for any researcher is that true research can never give a definitiveanswer about a phenomenon, since even the most well-known and basic principleis always subject to falsification.
The empirical work, addressed by this chapter, is concerned with assessing the effec-tiveness of BPMN model’s verification. An enhanced method is intended to be achieved
149
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.2. Choosing the Research Method
by adding a set of formalized well-formedness rules to the BPMN metamodel. We adopted,as research approach, the scientific method for assessing the improvements that one canachieve by using BPMN models’ verification with an enhanced version of the BPMNmetamodel.
8.2.2 Scientific Method’s Instantiation for BPMN Experiments
To customize the scientific method to the context of BPMN empirical studies, we followin the next sections the BPMN Empirical Study framework, which was inspired in experi-ment conduction and reporting guidelines frameworks [KPP+02, JP05, Ga08] used in theExperimental Software Engineering field. The activities carried out during the life cycleof an empirical study of BPMN models are depicted in Figure 8.1 and described next:
Figure 8.1: The BPMN empirical study framework
1. Definition – in this activity the addressed research problem is stated alongside withthe formulation of the research questions concerning expected results on BPMNmodels verification, the objectives of the experiments, as well as the context inwhich the experiments will be carried out (section 8.3).
2. Planning – this set of activities is about how the experiments will be performed,which involves a more detailed decisions concerning the context of the experi-ments. The formulation of a set of hypotheses under study is the basis for theexperiments’ design, in order to answer the initial research questions. We have alsoto make the elicitation of the set of independent and dependent variables that willbe used in the statistical tests, and the selection of subjects to participate in our ex-periments. The quantitative experiments are designed to filter out external factors,and to avoid bias in the results. The experiments’ instrumentation, as well as aprevious evaluation of their validity conclude the phase (sections 8.4.1 and 8.5.1).
3. Execution – consists in the instantiation of the previously established plan con-strained to the specific circumstances found in the actual experiments. In this phasedata is collected to gather empirical evidence (sections 8.4.2 and 8.5.2).
4. Data Analysis – the activities involved herein are the data set description and reduc-tion, as well as the hypotheses’ testing of the ones defined during each experimentplan (sections 8.4.3 and 8.5.3).
5. Results – since the overall research is constructed to allow comparability with re-sults from other researchers that would eventually repeat the experiments, the next
150
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.3. Empirical Studies’ Definition
set of activities consists is packaging the results so that they can be used by thecommunity. This involves documenting the whole experimental process, and dis-cuss the results achieved with the experiments, after the statistical analysis of theresults has been performed, focusing on aspects such as the results’ interpretation,the study’s limitations, the inferencing regarding the effectiveness of use of BPMNwell-formedness rules to the population of BPMN models, and the identification ofthe learned lessons (sections 8.4.4 and 8.5.4).
The next sections will illustrate the usage of BPMN Empirical Study framework in twoBPMN experiments concerning the empirical validation of BPMN well-formedness rules.
8.3 Empirical Studies’ Definition
After a previous work about the formalization and implementation of BPMN well-formednessrules using OCL (chapter 5), the motivation for the current empirical studies are to assessthe quality characteristic of correctness of BPMN models using those formalized rules.With this aim we will assess samples of BPMN models, from different sources (see sec-tion 7.3.2):
• BPMN models available from open repositories;• BPMN models produced by process modelers with different technical skills (pro-
cess analysts and process implementers).
Following the guidelines proposed in the BPMN Empirical Study framework overviewedin section 8.2.2, we address in this section the activities that are part of the experiments’definition, which consists upon specifying:
• the research problem – justifying the importance of the empirical studies;• the research objectives – highlighting the aims of the studies; and• the context definition of the study – defining the environment that constrains the
previous two (research problem and objectives).
8.3.1 Addressing Research Problems
After having previously established the topic of interest of this dissertation (section 1.2.1),we focus our study here in a more narrow research topic by tackling only one of theproblems placed (RPA). This will help us in the formulation of a set of research hypothesis(sections 8.4.1.2 and 8.5.1.2) to be tested (sections 8.4.3.3 and 8.5.3.3).
Our concern here is on assessing the quality of BPMN models produced using theBPMN standard. In other words, we want to investigate the effectiveness of BPMN stan-dard usage in different situations that lead to the production of BPMN models. So, thisresearch problem concerns the compliance of BPMN models with the quality character-istic of correctness. For quality compliance verification we use the set of BPMN well-formedness rules previously formalized in chapter 5. The samples came from (see sec-tion 7.3.2) (i) repositories of publicly available BPMN models; and (ii) BPMN models
151
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.3. Empirical Studies’ Definition
produced by process modelers from different technical backgrounds.
From the attained results we expect to justify the benefits of incorporating well-formednessrules within the BPMN metamodel. We also expect substantiating the benefits that under-pin the modeling process with tools amenable of verification of BPMN well-formednessrules.
8.3.2 Addressing Research Questions and Objectives
By tackling the problem mentioned in the previous section, were raised several questions:
• Do the BPMN models hosted in open repositories and built by process modelingexperts comply with the correctness quality requirement?• Is there any association between errors found in BPMN models and the violation of
best-practices in process modeling recommended by practitioners?• Is there any difference, regarding the quality, of BPMN models produced by mod-
elers with different technical backgrounds (process analysts vs process implemen-tors)?
To be more systematic and rigorous addressing to the above mentioned problems, aswell as to precisely delimit the boundaries of our empirical study, we formulated a set ofresearch questions (RQ). These research questions are stated to guide the empirical studythrough a short rationale descriptive, and detail the research question RQA formulatedin general terms on chapter 1. More complex research questions (RQAi) were break downin partial questions (RQAij where j is a letter that identifies a research sub-question).
1. [RQA1]: Can BPMN expressiveness hamper BPMN models’ correctness?• [RQA1a]: Can the set of well-formedness rules derived in chapter 5 be more
effective uncovering rule violations of process models than the verificationmade by BPMN experts?• [RQA1b]: Is it likely that process modelers with different technical skills (pro-
cess analysis or process implementation oriented) can deliver BPMN modelswith different quality characteristics?• [RQA1c]: Is it likely that different organizations interpret differently the BPMN
standard and therefore deliver BPMN models with different quality character-istics?
2. [RQA2]: Is there any association between the rule violations and the usage of specificconstructs in BPMN models?
• [RQA2a]: More exception paths, since it means more abnormal situations todeal with by a process modeler, increase the probability of BPMN well-formednessfaults?
• [RQ2b]: More parallel paths, since it means more mental states [Mil56] to dealwith by a process modeler, increase the probability of BPMN well-formednessfaults?
152
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.3. Empirical Studies’ Definition
3. [RQA3]: Is there any association between BPMN well-formedness faults and nonconformance with best-practices advocated by process modeling experts?
The rest of this chapter is intended to search for answers for the research questionsintroduced in this section. The research questions will be refined into research goals(sections 8.4.1.1 and 8.5.1.1), which in turn, will lead to the specification of the researchhypotheses (sections 8.4.1.2 and 8.5.1.2).
To provide grounded answers to the aforementioned research questions, we analyzedsamples of BPMN models collected from disparate sources, with respect to the effectivecompliance with BPMN rules, as well as generally accepted best practices (section 8.4).We analyzed also a sample of process models produced by modelers with different tech-nical skills, either domain or IT oriented (section 8.5).
For the experiments’ definition we apply the GQM framework [BCR94], already in-troduced in section 6.3.2. The GQM model hierarchical structure starts from a top-levelgoal definition. This goal is then refined into several questions, that usually break downan issue into its major components. Each question is also refined into measures. The samemeasure can be used to answer different questions under the same goal. The frameworkis instantiated through a template with a certain number of items (namely study object,purpose, quality focus, viewpoint and environment), which helps delimiting the experiments’boundaries. This is done by focusing on the relevant goals, and determining for exam-ple the related entities to measure, the dependent and independent variables, and thehypotheses to be formulated.
The instantiation of the GQM framework for the empirical study of this chapter, ismade by establishing the research objective GA, as the GQM’s top-level goal, and statingthis goal in the terms prescribed by the above mentioned template:
Analyze BPMN modelsfor the purpose of characterizing the usage of BPMN concerning well-formednessruleswith respect to the assessment of the quality characteristic of correctness,from the point of view of BPMN modelers (process analysts and process imple-menters),in the context of experimental studiesconstrained by the specificities of the multiple sources from which the samples werecollected.
8.3.3 Context Definition
The context of the BPMN experiments determines whether the experimental results aregeneralizable or not to a broader context. It is important to make the context explicit forthe sake of the comparability of results, since each experiment can have its own distinctcontext, with specific costs, benefits and risks.
The studies presented throughout this chapter, are developed in the context of BPMN
153
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.4. First Empirical Study
models’ verification, using two samples of BPMN models2:
• S1 – collected from public repositories managed by two BPMN tool providers (Ap-pendix D); and• S2 – collected from an experiment carried out with students of two degrees, in
a course on BPMN modeling taught at ESTSetúbal3 academic institution, using amodeling exercise detailed in Appendixes F and G.
Although the conclusions of our empirical studies can be generalizable to other BPMNmodeling contexts, caution must be taken and further research is required to confirm theattained results, before the generalization of the results.
8.4 First Empirical Study
In the following sections we apply the guidelines for experiments’ reporting proposed in[JP05] to our first empirical study. For the sake of understandability, we detail the rationalof each step of the reporting framework.
8.4.1 Empirical Study Planning
In Empirical Studies’ Definition (section 8.3), we already justified the need for BPMN ex-periments. This section is concerned with how the experiments will be performed. Beforeactually conducting the experiments in section Empirical Study Execution (section 8.4.2),we will characterize here with more detail the environment in which we will pursuitexperiments.
This will be the basis for introducing in the following sections, the research goals (sec-tion 8.4.1.1), the hypotheses under analysis and the group of independent / dependentvariables that will be considered to assess the hypotheses (section 8.4.1.2), the criteria forselection of subjects participating in the experiment (section 8.4.1.3), the experiments’ de-sign and instrumentation (section 8.4.1.4), and an evaluation of the experiment’s validity(section 8.4.1.6).
The outcome of this section is an experimental design, i.e., a receipt for the exper-iments, which provides the information for their replication by other researchers, andalso to allow readers to evaluate the experiments’ internal validity.
8.4.1.1 Research Goals
The research objective GA outlined in section 8.3.2 is refined here as research goal. Weuse the same index for goal and research question for the sake of traceability betweenthem. So, the goal GA1 corresponds to research question RQA1, and so on. Below, wedecompose GA1 and GA2 into sub-goals.
2The samples used in this dissertation are available in http://sdrv.ms/1i451Cd3http://www.ips.pt/
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.4. First Empirical Study
• GA1 (wrt RQA1) – analyze BPMN models for the purpose of assessing constructs’usage with respect to compliance with BPMN well-formedness rules, from thepoint of view of the BPMN standard, in the context of empirical studies con-strained by the usage of samples of BPMN models.∗ GA1c (wrt RQA1c) – analyze BPMN models for the purpose of assessing providers
with respect to the quality characteristic of correctness of their models, from thepoint of view of the BPMN standard, in the context of an experimental studyconstrained by a set of BPMN models collected from open repositories.
• GA2 (wrt RQA2) – analyze BPMN models for the purpose of associate the usage ofBPMN constructs with respect to compliance with BPMN well-formedness rules,from the point of view of the BPMN standard, in the context of experimental studyconstrained by a set of BPMN models from open repositories.∗ GA2a (wrt RQA2a) – analyze BPMN models for the purpose of associate the us-
age of exception paths with respect to compliance with BPMN well-formednessrules, from the point of view of the BPMN standard, in the context of experi-mental study constrained by a set of BPMN models from open repositories.
∗ GA2b (wrt RQA2b) – analyze BPMN models for the purpose of associate the usageof parallel paths with respect to compliance with BPMN well-formedness rules,from the point of view of the BPMN standard, in the context of experimentalstudy constrained by a set of BPMN models from open repositories.
• GA3 (wrt RQA3) – analyze BPMN models for the purpose of associate BPMN stan-dard’s rule violations and non conformance with best-practices advocate by prac-titioners with respect to the positive association among them, from the point ofview of BPMN modeler, in the context of an experimental study constrained by aset of BPMN models from open repositories.
8.4.1.2 Hypotheses and Variables
8.4.1.2.1 Hypotheses
The hypotheses presented in this section suggest explanations for the general phenomenonof BPMN models’ correctness. The null hypothesis represents the explanation for the phe-nomenon that we want to challenge, i.e., is the one we will try to disprove. Conversely,we will provide an alternative hypothesis, our research hypothesis that traduces the exis-tence of a factor which contributes for the correctness of BPMN models. By testing thehypotheses to disprove the null hypothesis, we expect to contribute for the explanationof BPMN models’ correctness.
The goals settled in section 8.4.1.1 guided us through the derivation of eight differenthypotheses. For each one of them, we formulate the null hypothesis (denoted by theindex 0), leaving the alternative hypothesis (index 1) to be inferred by negating the nullone.
155
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.4. First Empirical Study
• HA10 (wrt GA1) – BPMN models built by experts have no significant BPMN well-formedness rules violations.∗ HA1c0 (wrt GA1c) – The quality characteristic of correctness in BPMN models, has
no significant difference, regardless the organization who publishes the processmodel.
• HA20 (wrt GA2) – BPMN models well-formedness rules violations have no significantcorrelation with the number of constructs used.∗ HA2a0 (wrt GA2a) – BPMN models well-formedness rules violations have no signif-
icant correlation with the use of exception paths in BPMN models.∗ HA2b0 (wrt GA2b) – BPMN models well-formedness rules violations have no signif-
icant correlation with the use of parallel paths in BPMN models.• HA30 (wrt GA3) – The BPMN standard’s rule violations on process models, have
no significant correlation with BPMN best-practices’ violations found in the sameBPMN models.
8.4.1.2.2 Variables
A variable is a particular attribute or characteristic of an entity which is being observed,so can be measured and assume more than one of a set of qualitative (aka, categorical) orquantitative (aka, numeric) values. Examples of variables in our particular case are forinstance, the number of faults in BPMN models or the background of process modelers.
A factor (aka, independent variable), in any experiment or observational study, is aparticular type of variable that can be manipulated and influences the outcome of experi-mental study [Shu09].
The independent variable (the cause) is the variable which we would like to measure,while the dependent variable is the assumed effect that relies on the independent variable.Both type of variables can be stated in a hypothesis and should be explicitly tied to theresearch goals of the experiment. A well-designed experiment normally incorporates oneor two independent variables, and two or more dependent variables [Shu09].
For selecting the variables related with the hypotheses formulated in the previoussection, we rely in the GQM framework [BCR94], namely on the formulated researchgoals. The dependent variables’ elicitation is guided by the quality focus of each goal,which is preceded by with respect to sentence in each research goal’s formulation. Thedependent variables are the link between the data to be collected and research goals tobe achieved.
The independent and dependent variables used for each identified hypothesis arelisted in the Tables 8.1 and 8.2. Each dependent variable is specified in terms of theconstructs of the BPMN metamodel and formally defined through an OCL expressionwhich allows its computation. The definition of the variable refers the attributes to bemeasured, the computation rule to be applied, and the unit of measurement assigned.
156
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.4. First Empirical Study
Table 8.1: Independent and dependent variables for HAi hypotheses
Role Variable Name Description
Indep. SourceA nominal variable which identifies the sourceinstitution (BizAgi or Trisotech) of the BPMNmodel in the sample S1.
Indep. Total ElementsAn absolute scale variable representing the to-tal number of instances of modeling constructsused in the BPMN model.
Indep. Boundary EventsAn absolute scale variable that conveys thenumber of instances of type BoundaryEvent inthe BPMN model.
Indep. Parallel GatewaysAn absolute scale variable that conveys thenumber instances of type Gateway that generateparallel SequenceFlow in the BPMN model.
Dep. Total S NOK
An absolute scale variable that conveys thenumber of well-formedness rules violated inthe BPMN model, according to our automaticmodel checker. Each well-formedness rule isformulated in terms of an OCL invariant (seeAppendix E for the complete list of rules).
Dep. Total BP NOK
An absolute scale variable that conveys thenumber of best-practice rules violated in theBPMN model. Each best-practice rule is for-mulated in terms of an OCL invariant (see Ap-pendix E for the complete list of rules)
Table 8.2: Independent and dependent variables used by HAi hypotheses
Variable Name Hypotheses
Source HA1cTotal Elements HA2
Boundary Events HA2aParallel Gateways HA2b
Total S NOKHA1, HA1c, HA2, HA2a,HA2b, HA3
Total BP NOK HA3
157
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.4. First Empirical Study
8.4.1.3 Subjects Selection
In the pilot study regarding this section, we plan the experiment to ensure that it is carriedout properly so that the results can actually reflect the actual reality of BPMN modeling,regarding models verification.
The population under consideration corresponds to the BPMN models from all or-ganizations produced by actual process modelers. Since it is unfeasible to collect ran-domized samples of the whole population, we relied on convenience sample S1. Althoughits inference capabilities are reduced, this technique can be used for documenting thespecificities of subjects (BPMN artifacts).
Our sampling strategy is a combination of the simple organization (all subjects of thesample S1 are treated equally) with convenience sampling (subjects are chosen based ontheir easier availability). The implications of this choice will be discussed, in the nextsection 8.4.4, regarding Empirical Study Results.
We use as representation of the population, an accessible population of BPMN mod-els, instead of a theoretical population. The population is made available through openrepositories by two BPMN standard’s contributors (BizAgi and Trisotech) that are partof the Finalization Task Force Voting Members and voted the final version of the BPMN 2.0.So, both providers of BPMN models are actual experts in the BPMN usage.
The BPMN models in the accessible population are distributed into 8 different organi-zational functions (Human Resources, Finance, Administration, Research & Development,Production, Sales, Supply Chain, Services / Support), and 13 industries (Financial Ser-vices, Insurance, Health-care, Government (Public Sector), Manufacturing, Telecommu-nication, Energy/Utilities, Consulting / Service Providers, Transportation, Retail, Phar-maceutical, Hardware / Software, Communication). This set of BPMN models seems tobe representative of the theoretical population, for the purposes of this study, given itsbroad coverage of domains of usage of BPMN models.
Since the set of process models made available through both providers is subject tofrequent updates, we include in our sampling frame all the BPMN models listed on bothrepositories on a particular moment in time (June/30/2012 00:00 GMT). Thus, our sam-pling frame included the enumerated set of 59 BPMN models, which were the union ofthe set of BPMN models made available by BizAgi and Trisotech.
Finally, we actually draw from the sampling frame our convenience sample by:
• discarding from the sampling frame all the To-Be models when the correspondingAs-Is model was available. The justification is the big resemblance between the twoversions of the process, which makes the To-Be version redundant in the sample;• discarding the models with syntactical errors, i.e., models that do not complied
with the abstract syntax of BPMN, because we are interested in assessing well-formedness of BPMN models;• excluding conversation and choreography process models, and including simply
orchestration process models since the rules that were going to be verified were
158
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.4. First Empirical Study
only related to internal process models.
The convenience sample S1 rested with a set of 48 BPMN models collected from thetwo tool providers for the realization of a post-mortem observational study.
8.4.1.4 Experimental Design
The experimental design is the framework that gives direction and systematizes the re-search, namely by constraining the statistical techniques that can be followed for ana-lyzing collected data from the experiment. Hence, the design of an experiment affectsfindings’ achievement and is also critical for the validity of the results. Several facts mayinfluence the choice of experimental design, such as feasibility, time, cost, ethics, andmeasurement problems. Moreover, the hypotheses and variables elicited, also place re-strictions on the experiment designs to choose.
For our experiment’s design we prescribe the division of the sample S1 into a set ofgroups. Each of those groups receives a set of interventions (observations or treatments).The characterization of the experimental design, to be detailed in next section, was basedin the sequencing and synchronization of the interventions, the specification of their na-ture, as well as the group definition policy.
Given the previously described sample S1 and the nature of the phenomenon in study(quality of BPMN models), we use the following types of research:
• descriptive design, which aims to observe and to describe the correctness of BPMNmodels through a descriptive research;• correlational design, searching for relationships between variables through a rela-
tional research;• quasi-experimental design, aiming to determine the impact of a treatment in the
phenomenon through the use of a semi-experimental research.
For each experiment we use a notation [Tro06] that depicts the experiment design asa set of parallel sequences of letters (one sequence for each group). The meaning of eachletter is concerned with the assignment to group – random assignment (R), non-equivalentgroup (N) or assignment by cutoff (C) – or the intervention assigned to the group – observation(O), or treatment (X). The time moves in sequence from left to right, so, elements that arelisted on the left occur before elements that are listed on the right. The parallelism andsynchronization of interventions among groups are represented by the vertical alignmentof the correspondent letters.
Following the previous considerations, we describe below the design of a set of ex-periments involving the intervention on artifacts stored in open repositories, for a seriesof post-mortem observational studies.
For the HA1 hypothesis we use the convenience sample S1 as a single group. The ar-rangement means that there is no control group in this post-test only non-experimentaldesign. In practical terms we pack all the BPMN models collected from the repositories,apply the checking procedure of well-formedness rules and verify whether the number
159
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.4. First Empirical Study
of violations detected is statistically significant. It consists in a one-shot survey with asingle observation, which is the simplest form of non-experiment. According to [Tro06]this is also a valid design and the most adequate form of research for the descriptivestudy to address our goal GA1 and the corresponding research question RQA1.
The concise representation of the non-experiment is the following:
X O
For addressing HA1c hypothesis we use the provider property of the BPMN models. Bychecking the well-formedness rules compliance of each group one can conclude whetherthe source of the BPMN model is a relevant factor regarding the number of violationsdetected. The BPMN models have a set of properties, thus it seems natural to split thesample S1 using these properties. The most appropriate sampling technique to be usedin this case is the non-equivalent groups design technique. This design option is recom-mended in situations such as the one where a nominal property of the BPMN modelswas chosen for non-random assignment to the groups.
Thus, the qualitative property of the BPMN models’ provider (Bizagi or Trisotech) isthe discriminator for group assignment. This design is a post-test only non-equivalentgroups quasi-experiment because there are multiple groups subjected to measurement[Tro06]. The concise representation of the quasi-experiment is the following:
N1 X ON2 X O
In our hypotheses HA2, HA2a, HA2b and HA3, we will perform correlational studies forcorroborating existent relationships (associations) among BPMN models variables. Wewant namely, to establish relationships among BPMN correctness, embodied by well-formedness rules compliance, and specific constructs of BPMN models: (i) all instancesof the graphical modeling constructs in HA2; (ii) boundary events that control exceptionpaths in HA2a; (iii) the number of parallel paths in HA2b; and (iv) the compliance of mod-elers with BPMN best-practices in HA3.
In each of the correlational experiments we use 2 or more quantitative variables fromthe same group of BPMN models, to determine whether there is a relationship (or covari-ation) between the variables. Although correlation cannot prove a causal relationship, wecan however use it for supporting the previous statements regarding BPMN constructsand models’ correctness.
The statistical measure of correlation strength is the correlation coefficient (r), and ourhypotheses should test the non independence between the variables.
For testing our hypotheses we use the convenience sample S1 as a single group. Theconcise representation of the non-experiment is the following:
160
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.4. First Empirical Study
X O
8.4.1.5 Collection Procedure
The collection process planning consists in defining a protocol for gathering the exper-imental data in a most cost effective way, as well as ensuring the quality of data. Forthat purpose, it is established who (e.g. researcher, participants) will collect data, whereit will be done, and what procedures should be followed. It should also be ensured theavailability of subjects when data collection is scheduled to be performed.
We consider for collecting a sample of BPMN models an archival research in openrepositories. The advantages of archival research is that the design of the gathered BPMNmodels is not influenced by the researcher. On the other hand, we can access thosemodels freely, which renders the research less expensive. We also benefit from the largeamount of BPMN models available in the repositories, which allows the number of sub-jects in our convenience sample S1 to exceed 40 BPMN models, thus contributing for thevalidity of the research.
We already covered in section 7.3.2, albeit from a model-driven perspective, the pro-cess of collecting BPMN artifacts stored in open repositories. The process model in Figure7.11 details the activities supported by each tool and the role they play in data collection.
The BPMN models are collected by downloading them from providers’ repositoriesand storing them locally in a directory on the file system. 73% of the total of those BPMNmodels come from the Trisotech repository and the remainder from the Bizagi web site.The Bizagi Process Modeler tool helped converting the Trisotech’s process models, inVisio file format, to the Bizagi file format. We chose to have all BPMN models in Bizagifile format because this tool has an importer utility of Visio files and also an exporterutility to XPDL file format.
The BPMN metamodel – complemented by the well-formedness rules derived inchapter 5, as well as the measures specified in chapter 6 – is the basis for the verificationof BPMN models. Each BPMN model after being converted to XPDL format is used asinput for the instantiation of the BPMN metamodel. This allows the creation of the meta-objects, the meta-links among them, as well as the assignment of values to the attributesof the meta-objects related with the BPMN model instantiation. The well-formednessrule violations and measure observations from the instantiated process model are storedin a text file, for further processing and transformation into an SPSS data file. Finally, canbe carried out the statistical treatment required for testing the formulated hypotheses.
8.4.1.6 Analysis Procedure
The analysis approaches chosen for the experiment is dependent of the adopted experi-ment design (section 8.4.1.4), the variables earlier defined (section 8.4.1.2.2), and the re-search hypotheses to be tested (section 8.4.1.2.1). If required, more than one technique
161
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.4. First Empirical Study
may be assigned to each one of the research hypotheses, so the results can be assessedlater. Furthermore, each hypothesis may be analyzed with a different technique, namelyif the variables involved in the hypothesis are different from the ones being used in otherhypothesis to be tested.
We do not discuss in this dissertation the applicable tests in detail. However, the scaleand level of measurement of the variables used in the statistics tests constrain the choiceof the most suitable tests.
Some activities included in data analysis are the following:
• Descriptive statistics: For the dependent and independent variables, we compute,present and analyze a set of descriptive statistics of the variables’ distribution (e.g.the mean, standard deviation, the minimum value, the maximum value, the skewness andthe kurtosis). The descriptive statistics provide a first glimpse on data that will bedetailed in subsequent analysis.• Data set reduction: the existence in the data distribution, of outliers and extreme val-
ues can modify the relations between dependent and independent variables. Out-liers and extreme values should be removed from the analysis when their presencebiases the analysis. Therefore, before proceeding with further tests, one must de-termine whether these values occur in the data distribution.• Normality tests: Before statistical tests be done in order to verify hypotheses, we
have to check the data’s distribution through the Kolmogorov-Smirnov and theShapiro-Wilk tests. This is important, so we can select the appropriate statisticaltests for data. In particular, normality tests allow the researcher to decide whetherto use parametric tests, or nonparametric tests. The former are generally more pow-erful than the latter. However, they require a known data distribution. The lattershould be used when data does not comply with a normal distribution. Anyway,further tests to the data may be required, to ensure the meaningfulness of the usedstatistical tests. The mentioned tests are pivotal for deciding the adequate statisticsfor hypothesis, given the distribution’s characteristics of the sample.The null hypothesis for the normality tests states the absence of statistically signif-icant difference between the observed accumulated distribution and the one of thetheoretical normal distribution. Considering the confidence interval of 95% in bothtests, the normality hypothesis should be rejected when the significance is less than0.05. Using different words it can be said that if the variable’s normality test has asignificance level (p-value) greater than 0.05, one cannot reject the variables’ distribu-tion normality, with a confidence level of 95%.• Correlation analysis: A kind of tests that allow verifying whether exists a statistically
significant association between independent and dependent variables. If yes, thedependent variable can be regarded as a clue of the independent one. This rela-tionship should also be further explored. Otherwise, as one can conclude on theabsence of association, the hypothesis under scrutiny can be rejected.
162
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.4. First Empirical Study
• Hypothesis testing: Depending on the sample’s distribution, a parametric or a non-parametric test is performed to check for statistically significant differences amonggroups of observations.
We will detail each of these activities in section 8.4.3. We will also provide specificinformation regarding particular statistic tests, emphasizing their interpretation whenthey are used.
8.4.1.7 Instrumentation
The instrumentation process is concerned with the researcher’s definition of the artifactsthat will be used during the experiments. The instrumentation also concerns guidelinesdefinition, as well as tools that will support the measurements within the experiment.Training material distributed to the participants, as part of the experiment, should alsobe pointed out. The aim is to make clear the logistics needed for data collection in case ofexperiments’ replication.
In the case of the current dissertation we are dealing with BPMN models correctnessexperiments. So, the list of BPMN well-formedness rules that are going to be used inBPMN models’ verification (Appendix E), either manually or automatically, is one of theartifacts that must be considered.
The instrumentation of the study also requires off-the-shelf tools as well as custommade tools, developed for the purposes of this work. Each tool is used independentlyso, the experiments’ environment follows the pipes and filters [BCK03] architectural style.Below we detail the role of each tool.
• The Enterprise Architect is a graphical editor of BPMN models.• The USE tool allows the specification models to use elements of UML class dia-
grams, enriched with expressions in OCL to specify both integrity and constraints.• The Eclipse is an IDE used here for building ATL transformations and a Java appli-
cation for querying the Enterprise Architect repository.• The Bizagi Process Modeler is a graphical editor of BPMN models, an importer of
BPMN models in Visio format and a transformer to XPDL format.• The SPSS is the statistical tool for processing and analysis of collected data. The tool
is used to perform all the statistical tests of hypotheses. By scripting and savingthe sequence of commands used for performing data analysis, it also allows theautomation and replication of the whole statistical treatment.• The BPMN2USE is a transformer that takes as input BPMN graphical models pro-
duced with the Enterprise Architect tool and instantiates the BPMN metamodelusing the USE concrete syntax.• The XPDL2USE is a transformer developed for parsing XPDL files and create in-
stances of the BPMN metamodel, in the USE concrete syntax.• The JUSE-JUnit is a Java facade for the USE tool used for well-formedness rules test-
ing and debugging. After each rule is codified, added to the BPMN metamodel and
163
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.4. First Empirical Study
syntactically validated, BPMN model snippets were modeled to test the correctnessof the rule.
8.4.2 Empirical Study Execution
In this phase, the Empirical Study Planning is instantiated. Since the plan instantiation,due to local constraints, can be done in several ways, it is important to document thespecificities of a particular experiment, beyond the previously defined in the experimentplan. For instance, the number of artifacts/subjects, as well as their source/backgroundmay differ, from one to another experiment.
An experiment is typically carried out by manipulating the independent variable.The effect we are interested in, the dependent variables, can then be measured throughthe generated data set, which allows the analysis by statistical means, namely by thehypotheses testing.
8.4.2.1 Sample
For the experimental work, we downloaded the BPMN models hosted in open repos-itories, in order that experiments could be conducted off-line. Although belonging toparticular organizations, the BPMN models are made freely available and without anyrestrictions concerning the usage for academic studies. So, no data clearance need to beobtained for data collection and treatment.
8.4.2.2 Preparation
The preparation of the experiment with the sample S1 from the open repositories con-sisted on gathering and developing the required computational tools. This included, asmentioned before, off-the-shelf tools (Enterprise Architect, Bizagi Process Modeler, Vi-sio, USE, JUSE-JUnit and SPSS), as well as the development of specific tools (XPDL2USE,BPMN2USE), to be part of the chosen pipe and filter architecture. A pilot study was con-ducted on a small sample of BPMN models, to test our process with actual data. Theperformance of the tools was not a requirement, given the relative small amount of sam-ple data. Once the tools interoperability among off-the-shelf and developed tools wasensured, the data collection was performed.
8.4.2.3 Data Collection
The data collection consists in the actual execution of the experiments. We recorded in-formation regarding the problems detected so that the experiments can be improved infurther replications.
In this observational study, our data collection’s process consisted in the download ofa sample of BPMN models available in open repositories. The main constraint in this taskwas the Bizagi Process Modeler tool that was used to convert the process models from
164
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.4. First Empirical Study
Trisotech in Visio file format (75% of the total sample) to the Bizagi file format, since thetool allows Visio file importing and XPDL file exporting. However, the Visio importerdid a clumsy job, so most of the Visio format process models had to be remade in theBizagi IDE and visually checked with the original. The XPDL exporter, on the otherhand, accomplished its job, notwithstanding two main drawbacks: (1) through XPDLmetadata it was not possible to locate flow nodes in lanes; (2) since BizAgi did not allowdepicting participants and lanes in sub-processes, if it was required, the workaroundwas to depict the sub-process as call activity; the lateral effect of this option was to dealwith an additional XPDL file for each sub-process which had to be merged into the mainprocess model file.
8.4.3 Empirical Study Data Analysis
After data have been collected through sample S1, we analyzed this sample. This processis detailed in next sections and involves the description of data sets (section 8.4.3.1), con-sideration regarding its reduction (section 8.4.3.2), as well as the testing of the hypotheses(section 8.4.3.3) defined during the experiments planning. By following these steps, weare instantiating the Analysis Procedure activity, part of the Empirical Study Planning.
8.4.3.1 Data Description
Data description helps understanding the gathered sample. Thus, a detailed data de-scription of the variables collected in our samples is presented in Table 8.3.
The raw data collected on our sample S1 have more data than the one needed for ouranalysis. To be able to statistically analyze relevant data, we filtered the raw data into datasets. Each data set became a collection of data ready for statistical analysis and inferencethat we used in our statistical tool (SPSS).
In the following sections we present for the sample S1, the descriptive statistics ofthe relevant variables for a previously formulated hypothesis. For each variable, wepresent the number of cases (N), the Mean, the Median, the Mode, the standard deviation(Std.Dev.), the Skewness, the Kurtosis, the Minimum value, and the Sum. The Kolmogorov-Smirnov with Lilliefors correction and the Shapiro-Wilk normality tests are also present. Thelatter is used mainly with smaller samples, and is presented here for the sake of confir-mation. We present the following statistics of those tests: the test statistic (Statistic), thenumber of degrees of freedom (df ) and the test significance (Sig.).
We begin exploring the sample by describing in Table 8.3 each of the variables inthe data set. Table 8.4 summarizes the results attained by checking the business processmodels publicly available. As can be seen, only 53.6% of the models were in conformancewith specification rules that are part of the BPMN standard. If furthermore, we had morestrict requirements, by imposing the conformance with best-practices modeling rules, thepercentage of models that would comply with these rules, would be drastically reducedto only 3.6%. These results underline the effectiveness and importance of OCL rules
165
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.4. First Empirical Study
embedded in the BPMN metamodel to attain correctness in business process models.
Table 8.3: Description of variables of the sample S1
Variable Description
Source The company that produced the modelModel BPMN model name, related with its domain
<999 OCLInvName>
133 variables related to distinct BPMN well-formednessand best-practice rules (implemented as a OCL invariant).The value of the variable is 1 if a violation to this particularrule was detected in the model and 0 otherwise
Total OKNumber of distinct BPMN well-formedness and best-practice rules that the BPMN model conforms to
Total NOKNumber of distinct BPMN well-formedness and best-practice rules violations
Total BP NOK Number of distinct best practices rules violationsTotal S NOK Number of distinct BPMN well-formedness rules violationsDistinct Elements Total number of distinct modeling elements usedDistinct Events Total number of distinct instances of event types used
Total CoveragePercentage of the available modeling constructs used by thecurrent BPMN model (Figure 3.2)
Event CoveragePercentage of the available BPMN event types used by thecurrent BPMN model (Figure 3.3)
<BPMN Construct>110 variables, each one has the number of times the corre-sponding modeling construct was used in the BPMN model
Activities Total number of instances of the Activity modeling elementEvents Total number of instances of the Event modeling elementGateways Total number of instances of the Gateway modeling element
ParalellGatewaysTotal number of instances of Gateway used in the BPMNmodel that have parallel SequenceFlow
Boundary EventsTotal number of instances of the BoundaryEvent modelingelement
Total Elements Total number of instances of modeling elements
In Figure 8.2, a series of Pareto’s diagrams highlight the usage of different kind ofelements in BPMN models. The shaded shape denotes the number of times a modelingelement was used in the BPMN models, starting with the most used elements closestto the origin till the scarcely used, on the right side. In the sample: (a) a small set ofthe BPMN constructs (20%) represent 80% of the elements used in the sample’s BPMNdiagrams; (b) 80% of all sort of modeling violations (well-formedness and best-practice)are due to a set of near 40% of modeling rules. For these figures contributes; (c) theviolation of almost 55% of the well-formedness rules; as well as (d) the violation of 40%of the best-practice rules.
The information depicted in Figure 8.2 allows us to conclude that in spite of the largeset of BPMN constructs (Figure 3.2), only a small set (20%) is actually used by BPMNmodelers. Furthermore, even models using a small subset of elements of the BPMNspecification, are error prone.
166
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.4. First Empirical Study
Table 8.4: Percentage of cases per number of rules violations
Figure 8.2: Pareto diagrams for BPMN elements and rules
167
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.4. First Empirical Study
So, one could expect to decrease the learning curve of new users on the notation, byfocusing on the correct usage of these small set of modeling elements. Indeed, this is theapproach to achieve quality upon models followed namely by the DSLs community. Itconsists in a strategy of reducing the number of constructs in the metamodel, aiming toattain correctness by construction of models [BAGaB11].
The same rational can be applied to well-formedness and best-practice rules to beprimarily taught to new users. Since they are the major source of BPMN models faults,this would probably mitigate the problem. From a total of 102 formalized rules of theBPMN standard, the sample S1 reported the violation of only 15 (15%). Regarding thebest-practice rules, we found errors in 19 (61%) out of 31 rules.
Tools that implement the BPMN metamodel with embedded well-formedness rulesare expected to contribute to a reduced learning curve of the modeling language for users,since these tools would assist the building of business process models. However, empir-ical studies should also be done to corroborate this conjecture.
Looking for differences, concerning BPMN elements’ usage, one can see in Figure8.3 that on average models from Trisotech are more concerned with the data perspective(ItemAwareElement and DataAssociation) in process modeling, while Bizagi’s models useon average more role assignment (Lane) and control-flow (Gateway) constructs.
Figure 8.3: Radar diagram depicting the BPMN elements’ usage by Source
168
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.4. First Empirical Study
In Table 8.5 we summarized the descriptive statistics for the dependent variables con-sidered for testing the hypotheses related with the sample S1. The normality tests ofthose variables are also presented (Table 8.6). Next we discuss these statistics for eachhypothesis.
Total S NOK 0.3 48 0.000 0.727 48 0.000Total Elements 0.175 48 0.001 0.704 48 0.000ParalellGateways 0.236 48 0.000 0.784 48 0.000Bound-ary Events
0.468 48 0.000 0.472 48 0.000
Total BP NOK 0.118 48 0.09 0.969 48 0.237
a. Lilliefors Significance Correction
8.4.3.1.1 HA1
In the HA1 hypothesis we want to test whether BPMN models collected from open repos-itories and built by experts are correct, i.e., whether they contain or not statistically signif-icant violations to the BPMN well-formedness rules.
Our dependent variable is Total S NOK. Its positive skewness (value 1.746 on Table8.5) indicates an asymmetric distribution, with a higher frequency of lower values. Thevalue of the kurtosis (value 3.19 on Table 8.5), allows us to say that the distribution is alsoleptokurtic, with higher probability of values being close to the mean than in a normaldistribution.
Both the skewness and the kurtosis of the distribution provide a clue on the non-normality of the variable’s data. Table 8.6 presents the results of two tests which confirm
169
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.4. First Empirical Study
the non-normality of the Total S NOK variable: the Kolmogorov-Smirnov with the Lillieforscorrection and the Shapiro-Wilk’s normality tests. The null hypothesis for each test is thatthe sample comes from a Gaussian (normal) distribution. The alternative hypothesisassumes that the sample comes from a non-normal distribution. With a p-value < 0.01 inboth tests for Total S NOK variable, we cannot assume the normal distribution of data.
We should definitely choose a non-parametric test to assess hypothesis HA1, sincewe have enough statistical confidence that the population is far from being normally dis-tributed. The One sample Wilcoxon signed-rank test is the non-parametric statistical used totest whether a sample mean differs significantly from a hypothesized value. We will useit to test whether the mean of violations to the BPMN well-formedness rules on processmodels built by experts is equal to zero.
8.4.3.1.2 HA1c
In the HA1c hypothesis we want to test whether the quality characteristic of correctnessin BPMN models is statistically significantly different in models published by distinct or-ganizations. In this hypothesis it is also used as dependent variable Total S NOK. FromHA1 hypothesis we already know that the variable is right-skewed and leptokurtic. So,we have also to use non-parametric tests to assess hypothesis HA1c.
The Wilcoxon-Mann-Whitney test for two unpaired samples is the non-parametric statis-tical test that we use for verify whether the means of two samples differ significantly.The samples that are considered are the two distinct sets of BPMN models that camefrom Bizagi and Trisotech repositories. We will test whether the mean of violations to theBPMN well-formedness rules in each set differ significantly.
8.4.3.1.3 HA2
In the HA2 hypothesis we want to test whether the BPMN models correctness is associatedwith the amount of graphical constructs used in models.
Besides the dependent variable Total S NOK we also have the variable Total Elements.The positive right-skewed and leptokurtic indicates the the non-normality of this variable(values of skewness/kurtosis of 3.195/15.315 on Table 8.5). Table 8.6 presents the resultsof normality tests that confirms the significance of the test and thus we cannot assume anormal distribution of Total Elements.
The non-parametric tests of Spearman’s rank correlation coefficient and Kendall rank cor-relation coefficient are used to assess hypothesis HA2. Spearman’s coefficient assesses howwell the relationship between two variables can be described using a monotonic function.Kendall’s coefficient is a statistic used to measure the association between two measuredquantities. We will test whether exists an association between the number of graphicalconstructs used in process models and the violations to the BPMN well-formedness rules.
170
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.4. First Empirical Study
8.4.3.1.4 HA2a
In the HA2a hypothesis we want to test whether the correctness of BPMN models is as-sociated to the number of exception paths represented by instances of BoundaryEventconveyed by Boundary Events variable.
So, besides the dependent variable Total S NOK, we have also the variable Bound-ary Events, which is right-skewed (value 2.843 on Table 8.5) with a value of the kurtosis(value 8.064 on Table 8.5), indicating that the distribution is leptokurtic.
The non-normality of data distribution of this variable is confirmed in Table 8.6 thatpresents the value of 0.00 in Sig. columns of Boundary Events variable indicating thesignificance of the test.
The non-parametric tests of Spearman’s rank correlation coefficient and Kendall rank corre-lation coefficient are used to assess hypothesis HA2a. We will test whether exists an associ-ation between the number of graphical constructs that trigger exception paths in processmodels and the violations to the BPMN well-formedness rules.
8.4.3.1.5 HA2b
In the HA2b hypothesis we want to test whether the BPMN models correctness is associ-ated to the number of instances of control-flow elements in BPMN models with outgoingparallel paths.
Besides the dependent variable Total S NOK, we have also the variable ParalellGate-ways. It is right-skewed (value 1.86 on Table 8.5) and has a kurtosis (value 4.171 on Table8.5) which indicates a leptokurtic distribution.
The distribution indicates the non-normality of this variable. The value of 0.00 in Sig.columns of ParalellGateways variable confirms the significance of the test and thus thesample is not from a normal distribution.
The non-parametric tests of Spearman’s rank correlation coefficient and Kendall rank cor-relation coefficient are used to assess hypothesis HA2b. We will test whether exists a as-sociation between the number of graphical constructs that trigger several parallel pathsin process models – i.e. multiple mental states [Mil56] to process modelers – and theviolations to the BPMN well-formedness rules.
8.4.3.1.6 HA3
In the HA3 hypothesis we want to test whether the BPMN standard’s rule violations onprocess models is associated with BPMN best-practices’ violations found in the sameBPMN models. The former is conveyed by Total S NOK and the latter by the variableTotal BP NOK. This variable is almost symmetric (value 0.012 on Table 8.5) and the valueof the kurtosis (value -0.28 on Table 8.5), indicates that the distribution being platykurtic,it is almost mesokurtic.
171
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.4. First Empirical Study
Table 8.6 presents the results of the normality tests. The value of Sig. column ofTotal BP NOK variable confirms the non-significance of the test and thus we can assumethat the sample comes from a normal distribution.
The non-parametric tests of Spearman’s rank correlation coefficient and Kendall rank cor-relation coefficient are used to assess hypothesis HA3. We will test whether exists an associ-ation between the BPMN best-practice rules violations found in process models and theviolations to the BPMN well-formedness rules.
8.4.3.2 Data Set Reduction
The data description allows the detection of uncommon cases such as incorrect values,outliers, or extreme values. Statistical outliers are data points numerically distant fromthe rest of the values. To avoid the insertion of bias on the subsequent analysis to be per-formed, one should consider the removal of uncommon cases before hypotheses testing.
Statistical outliers are however common in distributions, such as the one of most ofthe variables we are dealing with in our sample, which do not follow the traditionalnormal distribution. Therefore, in distributions with a heavy tail, when the assumptionof a normal distribution do not hold, the presence of statistical outliers can be seen asmore common.
In the case of our sample S1, it was not performed any data set reduction. We chosenot to remove outliers and extreme values in the sample, as this would mean a threat tothe validity of the achieved results.
8.4.3.3 Hypotheses Testing
From previous section 8.4.3.1 we know how data are organized, how many groups (pairedor unpaired) we have to consider, and whether data are drawn for a normal or non-normal population. We have also formulated in section 8.4.1.2.1, the null hypothesesH0, which describe the assumptions regarding the data sets. In this section, through hy-pothesis testing, we want to contrast each null hypothesis H0 against its correspondingalternative hypothesis H1.
Each null hypothesis establishes the absence of a claimed observable pattern in thedata set, so any found variations are due to chance. So, H0 is treated as valid unless theactual data contradicts it. Using a statistical tests we try to reject H0 against the alternativehypothesis which states that the variations observed are not coincidental. The statisticaltest is concerned only in testing the null hypothesis. One can only either reject the nullhypothesis, or not reject it. A statistical test can never prove the alternative hypothesis,since falsifiability demands that the hypothesis is never completely accepted, only whenthe null is rejected.
Through significance tests we decide whether the null hypothesis is supported or not.Statistical tests are the tools that allow us to conclude for the existence of statistical signifi-cance. A relevant statistical significance mean that there is statistical evidence, beyond the
172
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.4. First Empirical Study
mere observer’s judgment, to support scientific decision of rejecting the null hypothesis.
In hypothesis testing is always assumed a predefined level of significance, denotedby α. α represents an acceptable probability of wrongly rejecting the null hypothesis H0
when it is in fact true. It is also known as type I error or false positive determination. p-value of a statistical hypothesis test is the probability of attaining a test statistic value asextreme as or more extreme when the null hypothesis H0 is true.
One can also incur in the type II error or false negative determination, i.e., fail to reject thealternative hypothesis, although it is in fact false. The probability for making this error,called β, is often unknown. Type II errors are frequently due to small samples. Hence,the power of the test is defined as the probability of not committing a type II error, andshould preferably be as close as possible to 1.
The steps we followed to apply a statistical test are summarized below:
• Confirm the statements regarding null and alternative hypotheses to be tested, aswell as check whether the pre-conditions for the tests to be performed as stated insection 8.4.3.1. When comparing two groups, we must also distinguish betweenone-tail and two-tail tests. We choose a one-tail test when we must predict whichgroup will have the larger population parameter before we collect any data.• Define the significance level at 5%, corresponding to the value of 0.05 for α.• Obtain using the SPSS statistical tool the p-value, which is computed getting the
test’s statistics (S) and comparing it to the relevant critical values (CV). If the esti-mate point of a population parameter is P, with confidence interval [a, b] at confidencelevel C = 1 - α, then any value outside the interval [a, b] will be significantly differ-ent from P at significance level α, under the same distributional assumptions thatwere made to generate the confidence interval [MB10].• Decide to either fail to reject the null hypothesis or reject it in favor of the alternative
hypothesis. The decision rule can be made through two methods: (a) using therejection region approach one should reject the null hypothesis H0 if S > CV, andotherwise accept it; (b) using the p-value one should reject the null hypothesis H0 ifp ≤ α, and otherwise accept it. In this section we will use the latter approach.
The soundness of our hypotheses test should be verifiable by external observers read-ing experiment’s report in this dissertation. We try to achieve this through a descriptionof the results’ tests, their probability, degrees of freedom, direction, as well as test power.This level of detail is also essential so that comparisons could be meaningful when tryingto combine results from tests performed in independent experiments. A detailed presen-tation of results is also fundamental in order to support the results’ interpretation to bemade in section 8.4.4.1.
8.4.3.3.1 HA1
For testing HA1 we have measured the variable related with the number of faults in eachmodel (Total S NOK), with all BPMN models as one group. Since those BPMN models
173
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.4. First Empirical Study
were built by experts we would expect the faults’ mean to be equal to the hypotheticalvalue of zero. If not we would like to know whether the difference is statistically signifi-cant.
The One sample Wilcoxon signed-rank test is the non-parametric statistical test toverify whether the sample mean differs significantly from the hypothesized value of zero.
The test starts by ranking all the observations, irrespective of the sample they camefrom. Values are ranked from Negative Ranks to Ties. There are no Positive Ranks, since thenumber of violations in BPMN models is a non negative number. Table 8.7 summarizesthe information concerning the computed ranks.
The One sample Wilcoxon signed-rank test summarized in table 8.8 leads us to rejectthe null hypothesis. So, there is a high probability that the sample has as mean a valuedifferent from zero.
These results allow us to conclude that the BPMN models that came from open repositoriesand were built by BPMN experts are provided with significant well-formedness violations.
Table 8.7: Ranks for HA1
N Mean Rank Sum of Ranks
Error Free - Total S NOK Negative Ranks 22a 11.5 253Positive Ranks 0b 0 0
Ties 26c
Total 48
a. Error Free < Total S NOKb. Error Free > Total S NOKc. Error Free = Total S NOK
Table 8.8: Wilcoxon Signed Ranks test for the Total S NOK variable
Error Free - Total S NOK
Z -4.197a
Asymp. Sig. (2-tailed) 0.00
a. Based on positive ranks.
8.4.3.3.2 HA1c
The hypothesis HA1c concerns whether there are significant differences between the BPMNwell-formedness rules violations found in process models originated from different sources.We have to compare statistics of two unpaired groups of BPMN models and test whetherthe two groups means differ significantly. Our independent variable is Source, and weuse it to discriminate between the two groups (the BPMN models from Bizagi repositoryand the models coming from Trisotech).
174
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.4. First Empirical Study
Table 8.9: Ranks for HA1c
Source N Mean Rank Sum of Ranks
Total S NOK 1-Bizagi 15 32.1 481.52-Trisotech 33 21.05 694.5
Total 48
In order to compare the correctness of these BPMN models, we will perform theMann-Whitney U test (aka Wilcoxon-Mann-Whitney) to evaluate whether the two sam-ples come from the same population, by testing whether their means differ significantly.The test starts by ranking all the observations, regardless the sample they come from.Table 8.9 summarizes the information concerning the computed ranks.
Table 8.10: Mann-Whitney U test for the Total S NOK variablea
Total S NOK
Mann-Whitney U 133.5Wilcoxon W 694.5Z -2.794Asymp. Sig. (2-tailed) 0.005
a. Grouping Variable: Source
As we can see, 15 of the analyzed BPMN models come from the Bizagi repository, andthey have a lower mean rank. Note that the lowest ranks correspond to the highest val-ues. If the distributions come from the same sample, they should have equal probabilitydistributions. The Mann-Whitney test summarized in table 8.10 takes us to reject the nullhypothesis, which states that the two samples come from the same population. So, weconclude for an high probability of BPMN models being from different samples.
The test’s results are confirmed with a Two-Sample Kolmogorov-Smirnov test (table 8.11).The Kolmogorov-Smirnov statistic quantifies a distance between the empirical distribu-tion functions of the two groups. Under the null hypothesis it is stated that the two sam-ples are drawn from the same distribution. This test relies on the same rank classification(see table 8.9). Its significance confirms the results presented for the Mann-Whitney Utest.
These results indicate that the BPMN models produced by different sources have dif-ferent levels of correctness, i.e., the the BPMN models from different origins comply dif-ferently to the BPMN well-formedness rules established by BPMN standard.
In tables 8.12 and 8.13 we can collate the descriptive statistics of the sample by Source.One can conclude that although the number of the Bizagi BPMN models are less than theones provided by Trisotech, they contribute to the whole sample with a larger number ofwell-formedness rule violations, both in the total and by model.
175
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.4. First Empirical Study
Table 8.11: Two-Sample Kolmogorov-Smirnov test for the Total S NOK variablea
Total S NOK
Most Extreme Differences Absolute 0.400Positive 0.400
Negative 0.000Kolmogorov-Smirnov Z 1.285Exact Sig. (2-tailed) 0.014Point Probability 0.010
a. Grouping Variable: Source
Table 8.12: Descriptive statistics by Source (I)
Total S NOK
Source Mean Median N Std. Dev. Kurtosis Skewness1-Bizagi 1.53 1 15 1.506 0.695 1.0862-Trisotech 0.48 0 33 0.795 2.118 1.639Total 0.81 0 48 1.161 3.19 1.746
Table 8.13: Descriptive statistics by Source (II)
Total S NOK
Source Sum% of
Total Sum% of
Total N Min. Max. Range1-Bizagi 23 59.00% 31.30% 0 5 52-Trisotech 16 41.00% 68.80% 0 3 3Total 39 100.00% 100.00% 0 5 5
176
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.4. First Empirical Study
8.4.3.3.3 HA2, HA2a and HA2b
For testing HA2, HA2a, and HA2b, we should find whether the variable representing well-formedness rules violations in BPMN models (Total S NOK) is associated with the vari-able that represents the number of instances of BPMN constructs of a certain type (To-tal Elements, Boundary Events, Parallel Gateways). Basically, a correlational study looksat the strength of association between the variables. Correlation does not necessarilymean causality.
We start performing the correlation analysis, using the Spearman’s correlation test.For each of the studied hypotheses, we tested the correlations of the dependent variable(Total S NOK) with respect to each of the independent variables.
In the correlation test, the significance level showed is regarding the two-tailed asymp-totic significance. A p-value below 0.05 denotes a significance level, i.e., a statisticallysignificant correlation.
The HA2 hypothesis is concerned whether or not the number of constructs used (To-tal Elements) is correlated with the well-formedness rules violations in BPMN models.On the other hand, the HA2a hypothesis is concerned whether or not exception pathsin BPMN models (Boundary Events) are correlated with the well-formedness rules vi-olations in BPMN models. Finally, with HA2b hypothesis we try to measure whetheror not the number of constructs with outgoing parallel paths in BPMN models (Paral-lel Gateways) is correlated with the well-formedness rules violations in BPMN models.
The correlation analysis for the three hypotheses is summarized in table 8.14. Wecan observe significant correlations between our independent and dependent variables forhypothesis HA2 and HA2b and no significant correlations for hypothesis HA2a.
Table 8.14: Spearman’s rho for HA2, HA2a and HA2b
Correlations Total ElementsBoundary
EventsParallel
Gateways
N 48 48 48Total S NOK Correlation Coefficient .549** 0.141 .422**
Sig. (2-tailed) 0.000 0.339 0.003
**. Correlation is significant at the 0.01 level (2-tailed).
In table 8.15 is shown the strong positive relationship between all instances of FlowN-ode in the BPMN models (Total Elements) and the main instances of subtypes of FlowNode(Activity, Event, Gateway).
We can conclude that the overall number of constructs, and particularly the numberof constructs with outgoing parallel paths used in BPMN models, are associated to theobserved defect violations found in BPMN models. So, one can reject the hypothesis thatBPMN models well-formedness rules violations have no significant correlation with totalnumber of instances used in the BPMN models.
177
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.4. First Empirical Study
Table 8.15: Spearman’s rho for variables Activities, Events, Gateways
Activities Events Gateways
N 48 48 48Total Elements Correlation Coefficient 0.902** 0.725** 0.851**
Sig. (2-tailed) 0.00 0.00 0.00
**. Correlation is significant at the 0.01 level (2-tailed).
8.4.3.3.4 HA3
For testing HA3 we should find whether the dependent variable representing well-formednessrules violations in BPMN models (Total S NOK) is correlated to the independent variableTotal BP NOK.
The non-parametric test performed to validate the hypothesis HA3 is computed basedon the ranks of the values in the sample instead of the values themselves. The reason forusing ranks is due to the fact that one cannot rely on the assumption of normality in theanalysis of variance.
The Kendall rank correlation coefficient (aka Kendall’s tau coefficient) , is a statisticused to measure the association between two measured quantities. Specifically, it is ameasure of rank correlation, i.e., the similarity of the orderings of the data when rankedby each of the quantities.
We use the Kendall rank coefficient, under the null hypothesis, to test the indepen-dence of Total S NOK and Total BP NOK, assuming that the sampling distribution oftau has an expected value of zero. In Table 8.16 is shown that the test is statisticallysignificant, so there is a positive relationship between Total S NOK and Total BP NOK.
Table 8.16: Kendall’s tau b for HA3
Total S NOK Correlation Coefficient .430**Sig. (2-tailed) 0.000
N 48
**. Correlation is significant at the 0.01 level (2-tailed).
The previous test’s result is confirmed by the Spearman’s correlation test. The Spearmanrho test is also a measure of statistical dependence between two variables. It assesses howwell the relationship between two variables can be described using a monotonic function.In this test we use the same variables used in the Kendall rank coefficient.
In the correlation test, the significance level presented is the two-tailed asymptoticsignificance. As in the previous test, a p-value below 0.01 indicates a statistically signifi-cant correlation (Table 8.17).
So, one can reject the null hypothesis HA3 that BPMN standard’s rule violations onprocess models, have no significant correlation with BPMN best-practices’ violations found
178
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.4. First Empirical Study
Table 8.17: Spearman’s rho for HA3
Total BP NOK
Total S NOK Correlation Coefficient 0.522**Sig. (2-tailed) 0.000
N 48
**. Correlation is significant at the 0.01 level (2-tailed).
in the same BPMN models. Therefore, the violations of best-practices BPMN models areassociated to the observed well-formedness rules violations found in BPMN models.
8.4.4 Empirical Study Results
Since this experimental study is done, it is now crucial to package the achieved results soit can be used by the BPMN community. This involves substantiating the experimentalprocess, as previously discussed, which includes the discussion on the results achievedwith the experiments. This discussion is focused the interpretation of the results (sec-tion 8.4.4.1), as well as the limitations of the study (section 8.4.4.2), the inferencing thatcan be made regarding the extent to which the study’s results are expected to hold forpopulation (section 8.4.4.3), and the identification of the learned lessons (section 8.4.4.4).
8.4.4.1 Interpretation
This section is concerned with the analysis of the tests’ outcomes, supported on the theorybeing assessed. When the tests do not substantiate the theoretical assumptions, we try toidentify the causes that lead to the failure. Below we go through each of the hypothesis,by presenting and discussing the outcome of the hypotheses testing.
• RA1 (wrt HA1) – Empirical evidence allows us to reject the hypothesis that BPMN ex-perts generally produce correct BPMN models, i.e., those that do not violate BPMNwell-formedness rules.This result is in line with studies [RIRG05, RRIG09] that pointed out the increasein complexity of the latest BPMN standard [BPM11] due to the introduction of sev-eral new constructs (see Figures 3.2 and 3.3). The relatively more expressiveness ofBPMN came at a price of more error proneness even by BPMN experts. The solu-tion seems to be the attachment to BPMN design tools of a well-formedness ruleschecker. The proposal for the well-formedness rules formalization is presented inchapter 5 and listed in Appendix E.• RA1c (wrt HA1c) – Empirical evidence allows us to reject the hypothesis that BPMN
models quality correctness is the same regardless of institution.Different institutions interpret and implement BPMN well-formedness rules differ-ently, since they are informally expressed in natural language in the specification[BPM11]. The formalization of BPMN well-formedness rules through a rigorous
179
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.4. First Empirical Study
language such as OCL, as done in chapter 5, could mitigate this situation, as wellas facilitate the enforcement of those rules by tool makers.• RA2 (wrt HA2) – Empirical evidence allows us to considerer the hypothesis that
BPMN models well-formedness rules violations are associated with the numberof constructs used.This scalability problem comes at no surprise. Indeed, an increase in the models’dimension, without the adequate support of automatic tools, seems to be difficultto deal with by process modelers who fail to comply with modeling rules, therebygenerating faulty BPMN models.• RA2a (wrt HA2a) – Empirical evidence do not allow us to reject the hypothesis that
BPMN models well-formedness rules violations are independent of the number ofconstructs used to deal with exception paths.There is no evidence that an increase in the number of constructs used for docu-menting exceptions to the happy path, could be associated to an increase in the num-ber of BPMN well-formedness rules violations. In other words, tackling abnormalsituations is independent of the number of rule violations.• RA2b (wrt HA2b) – Empirical evidence allows us to reject the hypothesis that BPMN
models well-formedness rules violations are not associated with the parallel pathsused.There is some evidence that an increase in the number of constructs from whichparallel paths are derived is associated to an increase in the number of BPMN well-formedness rules violations. In other words, an increase in the number of mentalstates in process modeling is associated to the number of rule violations. This find-ing is aligned with the one reported by Cardoso [CMNR06] regarding the CFCmeasure.• RA3 (wrt HA3) – Empirical evidence allows us to reject the hypothesis that BPMN
standard’s rule violations on process models are not associated with BPMN best-practices’ violations. In other words, compliance with modeling rules advocatedby practitioners are associated with well-formedness rules compliance in BPMNmodels.
8.4.4.2 Validity Threats
The threats’ identification aims to substantiate relevant weaknesses of the previously de-scribed empirical work. This analysis can be seen as an opportunity for a systematicapproach to the experimental work for identifying supplementary studies that can con-tribute to improve the BPMN modeling process [Ga08].
While packaging the experimental results, we evaluate the whole experimental pro-cess, namely activities such as sampling or experimental design selection, and try to iden-tify the potential threats to the validity of the results. Moreover, we discuss the measuresput in place to address those threats.
180
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.4. First Empirical Study
We will consider for categories of validity threats in this section the ones advocated by[Ga08] that considers four aspects to deal with: (1) internal validity which is related withthe validity of the study itself, namely with the causal effect under study; (2) externalvalidity concerns with the researcher’s ability for generalizing the results to the industry;(3) construct validity refers to the the experiment’s results generalization to the underlyingtheory; and (4) conclusion validity is related to the ability to derive adequate conclusionsregarding the relations between the performed treatment and the research’s outcome.
8.4.4.2.1 Internal Validity
Concerning the internal validity, we consider multiple group threats.
Multiple group threats result from the fact that multiple groups differently exposedto single group threats. An effect of reduction of the comparability of obtained resultsmay occur. For S1 we consider the following threats.
∗ History. This threat is concerned with uncontrolled events that are irrelevant for thehypotheses under test. However, after those events occur confounding effects may beintroduced on the outcome of performed tests. In S1 some of the BPMN models weredone using BPMN version 1.2 (85%) while others were related to version 2.0 (15%). Theevolution of the BPMN standard brought differences in the number of constructs andrules of the older version. However, for the analyzed models, applying the set of rulesin BPMN 2.0 (a superset of the rules already existent in version 1.2) did not increasethe number of faults in the BPMN models under scrutiny.∗ Instrumentation. Errors in the measurements, either systematic or random, result from
measurement instruments not working as precisely as they should. Systematic or ran-dom events may jeopardize the quality of the experiment’s collected and consequentlythe interpretation of the results. In S1 the instrumentation used with Trisotech andBizagi BPMN models was slightly different, as described in section 7.3.2. However, itis unlikely the introduction of some sort of differentiation among the BPMN models,since only the models’ transformation was different.∗ Selection. When sampling from the population, one may incur in the risk that the sub-
jects do not represent the whole population. In S1 we are only using BPMN modelswhich were made available through public repositories. This excludes all the BPMNmodels not available in the repositories’ list, at the time of the sample collection. Ourconvenient sampling process, also has the potential to introduce some sort of bias, evenif we were not aware of it. We expected to have mitigated that bias by using BPMNmodels with different sources, size, and domains.
8.4.4.2.2 External Validity
External validity is related to the suitability of results’ generalization, beyond the scopeof the experiment. We elicit the following potential source of threat:
181
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.4. First Empirical Study
∗ Time. It may be the case that future developments of BPMN modeling tools may renderthe observations reported here as obsolete. Care must be taken on extrapolations toother contexts.
8.4.4.2.3 Construct Validity
Regarding construct validity, we consider the following design threats.
Design threats are the result of difficulties for rigorously define the causes and effectsbeing tested. This may lead to measurements and treatments poorly chosen, since thetheoretical concepts intended to be tested are not truly understood. In our experimentalstudies the design threats regarding construct validity include:
∗ Inadequate pre-operational explanation of constructs. This threat occurs from poorly de-fined constructs used in the experiment. It becomes more difficult to analyze theexperimental results if the theory underpinning the experiment is not clear. Our ex-periments are supported in OCL constraints for a rigorous measurement of all of thewell-formedness rules violations in experiments, as well as other measures needed bythe experiments based on S1. We regard this as an effective mitigation of this potentialthreat.∗ Confounding constructs and level of constructs. Besides assessing a construct with respect
to its presence, one should measure its level. In S1 BPMN models correctness is beingmeasured indirectly using a specific mechanism (number of well-formedness rules vi-olations). Differentiated replicas of this study could use eventually other mechanisms(e.g. ratio of detected errors by number of modeling constructs used in the BPMNmodel).
8.4.4.2.4 Conclusion Validity
These are the kind of threats that are inherent to statistical tests’ usage:
∗ Reliability of treatment implementation. A lack of standardization in the implementationof the treatment can originate a confounding factor. This may be due to an inconsistentadministration of the treatment in different groups. These variations happen morelikely if different people apply the treatment. However, they can also occur with thesame person administer the treatment. In S1 the observational study was conductedby the author of the dissertation, using automated tools. This uniformity mitigates thispotential threat.
8.4.4.3 Inferences
Following the testing phase of the experimental study, should be made conclusions fromthe statistical tests results. Through inference we apply the conclusions that have beenobtained from the experimental study to the general population. A study or experiment
182
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.5. Second Empirical Study
needs to state and conclude something about general populations and not just about thesample that was studied.
The analysis accomplished in this observational study is expected to hold for sam-ples with larger number of BPMN models. So, we expect that some patterns of well-formedness rules violations to be observable with other BPMN models made publiclyavailable. It seems also plausible that these observations could also applied to the BPMNmodels kept internal to organizations. We expect that some properties to hold, namelythe ones regarding to: (1) the difficulty to cope with the expressiveness of BPMN con-structs in order to ensure BPMN models’ correctness, even by BPMN experts; (2) differentBPMN model producers have different conformance to BPMN rules unless BPMN well-formedness rules get formalized and implemented, by tool makers, through automaticmodel checkers; (3) a positive association among constructs usage and the faults detectedin models built without automatic model checkers; and (4) a positive association betweenbest-practices violations and non-conformance with BPMN well-formedness rules, sug-gesting process modelers should comply with BPMN modeling best-practices to attainbetter quality from BPMN modeling.
8.4.4.4 Lessons Learned
In this section, are discussed the lessons collected from the operationalization of the ex-perimental study. Rather than the results’ discussion, we address the implications ofthese results.
Ensuring data quality for the statistical analysis, was an important challenge to over-come. A large amount of time and effort was required for data collection through down-loading and transformation of BPMN models. After gathering BPMN models we wereable to feed them into our pipeline of applications to support our analysis. We used trans-formers for automating the repetitive tasks (e.g. BPMN2USE and XPDL2USE), namelyfor generating the meta-objects and meta-associations to instantiate the BPMN meta-model. Collecting the figures for statistical analysis after that instantiation was relativelyfast.
By conducting the experiment, some process issues that can be improved were noted,as well as challenges that can be tackled in future replications of this experiment. Wewould like to concentrate the study on clusters of rules, rather than carry out the rulesone by one.
8.5 Second Empirical Study
In the following sections we report our second empirical study. Since we already detailedthe rational of each step of the framework for experiments reporting [JP05] (section 8.4),we only present here the obtained results.
183
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.5. Second Empirical Study
8.5.1 Empirical Study Planning
This step is the basis for introducing in the following sections: the research goals (section8.5.1.1), the hypotheses under study, as well as the independent and dependent variablesto be used for hypotheses’ evaluation (section 8.5.1.2), the criteria for selection of subjectsparticipating in the experiment (section 8.5.1.3), the experiments’ design and instrumen-tation (section 8.5.1.4), and an evaluation of the experiment’s validity (section 8.5.1.6).
8.5.1.1 Research Goals
The research objective GA, outlined in section 8.3.2, is also refined here through two moresub-goals of the research goal GA1.∗ GA1a (wrt RQA1a) – analyze BPMN models for the purpose of assessing effectiveness
of automatic and manual BPMN models’ verification with respect to well-formednessrules, from the point of view of the BPMN standard, in the context of an experimentalstudy constrained by the intervention of surrogates of actual BPMN modelers.
∗ GA1b (wrt RQA1b) – analyze BPMN models for the purpose of assessing complianceof BPMN well-formedness rules by process analysts and process developers with re-spect to the quality characteristic of correctness, from the point of view of the BPMNstandard, in the context of an experimental study constrained by surrogates of actualBPMN process modelers.
8.5.1.2 Hypotheses and Variables
8.5.1.2.1 Hypotheses
The hypotheses presented in this section also intend to suggest explanations for the gen-eral phenomenon of BPMN models’ correctness. By testing these hypotheses we expectto contribute for the explanation of BPMN models’ correctness.
The goals settled in section 8.5.1.1 guided us through the derivation of two differenthypotheses.∗ HA1a0 (wrt GA1a) – The effectiveness of BPMN experts verifying BPMN well-formedness
rules, has no significant difference from the one provided by automatic model checking.∗ HA1b0 (wrt GA1b) – The quality characteristic of correctness in BPMN models, has no
significant difference, regardless the technical background of process modelers (processanalyst and process developer).
8.5.1.2.2 Variables
The independent and dependent variables used for each identified hypothesis are listedin the Tables 8.18 and 8.19. Each dependent variable is specified in terms of the constructsof the BPMN metamodel and formally defined through an OCL expression which allows
184
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.5. Second Empirical Study
its computation. The definition of the variable refers the attributes to be measured, thecomputation rule to be applied, and the unit of measurement assigned.
Table 8.18: Independent and dependent variables for HA1a and HA1b hypotheses
Role Variable Name Description
Indep. Modeler TypeA nominal variable which identifies the techni-cal background of the BPMN modeler (processanalyst or process implementer) in the sample.
Dep. Total A NOKAn absolute scale variable that conveys thesame as variable Total S NOK, however col-lect from a different sample.
Dep. Total M NOK
An absolute scale variable that conveys thenumber of well-formedness rules violated inthe BPMN model, according to a verificationmade by a panel of BPMN experts.
Table 8.19: Independent and dependent variables used by HA1a and HA1b hypotheses
Variable Name Hypotheses
Modeler Type HA1bTotal A NOK HA1a, HA1bTotal M NOK HA1a
8.5.1.3 Subjects Selection
Our sampling strategy is a combination of the simple organization (all subjects of thesample are treated equally) with convenience sampling (subjects are chosen based on theireasier availability). The implications of this choice will be discussed, in the next section8.5.4, regarding Empirical Study Results.
The study was conducted from March to June of 2012 at Escola Superior de Tecnologiade Setúbal (ESTSetúbal)4, a polytechnic institution. It included subjects attending twodegrees with distinct curricula. The participants were 1st cycle undergraduate studentsof Bologna’s Program of Studies, in the second semester of their academic year. Thecharacteristics of the two graduations are described next.
• Technology and Industrial Management5 – this degree includes courses with topicssuch as industrial processes, and information systems (with process modeling us-ing BPMN).Given the more business and industry oriented nature of the topics taught, wechose the students from the second year of this degree as process analysts surrogates;
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.5. Second Empirical Study
• Informatics Engineering6 – in this degree are taught computer science topics, suchas procedural and object-oriented programming languages, systems analysis anddesign (using UML) and process modeling (using BPMN).The prevalence of implementation topics in this degree, led us to choose studentsfrom the third year of this graduation as process implementers surrogates.
The students of both degrees attended a process modeling course in the same semester,which was a good opportunity to enroll all of them in the same BPMN modeling exer-cise. The context of the course was a learning environment with students having a firstcontact with the BPMN. The students were provided, through an e-learning platform,course materials (course’s notes and tools usage) regarding BPMN modeling, as well asthe exercise’s description, one of the course assignments.
The number of participants was 51 (18 process developers and 33 process analysts).The experiment had the two following phases.
• Training – this phase was mainly a period of familiarization with the notation.Therefore, the output produced in this phase was not used in data analysis. Was de-livered detailed information to the students, about the domain of financial servicesprovisioning using Automated Teller Machines (ATMs) (see Appendix F). The pro-cess modelers were asked to read, understand and model the exercise using BPMN.A deadline of one month was given for the accomplishment of this preliminary as-signment.The main intent of this phase was that students could: (1) be acquainted with theprocess modeling tasks when performing a realistic modeling exercise; (2) prac-tice the application of the process modeling and BPMN concepts taught in regularclasses; (3) use a BPMN tool capable of basic syntactical checking of BPMN models,hence the tool Sparx Enterprise Architect (it was provided a video tutorial about itsusage7).• Experiment – the actual experiment took place in a laboratory class with a time frame
of two hours. At the beginning, the instructor explained to the participants whatthey were expected to do during that experiment, namely: (1) read the business pro-cess of Withdraw Cash (Appendix G), which was a shorter part of the overall casepresented in the previous phase; (2) complete the BPMN model and store it in aprovided repository of the Enterprise Architect tool. As part of the provided repos-itory, a baseline of the solution was delivered to the students, in order to delimit thescope and variability of their solutions.
With this experiment we intended, as primary goal, to evaluate the quality of theBPMN models produced by surrogates of professional process modelers with two dis-tinct modeling profiles (process analysts and process implementers), whom were pro-posed to solve the same modeling exercise using the BPMN standard.
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.5. Second Empirical Study
8.5.1.4 Experimental Design
This section describes the design of a set of experiments involving the intervention onartifacts produced by undergraduate students, as surrogates of process modelers withdifferent backgrounds.
In this non-experimental study regarding the HA1a hypothesis, we chose a within-groups design. This is a type of experimental design in which all subjects are exposed toevery treatment. In our case the subjects are the BPMN models, which are viewed as asingle group and are subject to more than one treatment: (i) manual verification by BPMNexperts; and (ii) automatic verification through a model checker. Each treatment refersto a level of the independent variable. We checked the well-formedness rules violations(dependent variable) scored by each type of verification method (independent variable)in order to answer the question regarding the effectiveness of the methods, i. e., whichmethod is more effective discovering BPMN well-formedness rule violations.
There are two fundamental advantages of the within-groups design: (1) increasedstatistical power; and (2) reduction in error variance associated with BPMN models dif-ferences due to the verification method.
• A fundamental inferential statistics principle is that, as the number of participantsincreases, statistical power increases, and therefore the probability of β error de-creases (the probability of not finding an effect when one truly exists). By using awithin-groups design we have increased the number of sample cases relative to abetween-groups design.• The reduction in error variance is due to the fact that much of the error variance in a
between-groups’ design is due to the fact that, even though we randomly assignedsubjects to groups, the two groups may differ with regard to important individualdifference factors that effect the dependent variable. With within-groups designs,the conditions are always exactly equivalent with respect to individual differencevariables since the BPMN models are the same in the different conditions. So, in theparticular case of HA1a hypothesis, any factor that may affect performance on thedependent variable (detection of well-formedness rules violations) such as the pre-vious modeling expertise of process modelers, or previous knowledge of the mod-eled domain by process modelers, will be exactly the same for the two conditions,because they are the exact same group of BPMN models in the two conditions.
The concise representation of the non-experiment with within-groups design address-ing HA1a hypothesis is the following:
X1 O X2 O
Quasi-experimental design involves selecting groups, upon which a variable is tested,without any random pre-selection process. The researcher treats the situation as an ex-periment even though strictly, by design, it is not. The independent variable may not bemanipulated, treatment and control groups may not be randomized or matched, or there
187
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.5. Second Empirical Study
may be no control group, so the researcher is limited in drawing conclusions [Aca13].
Non-equivalent groups are often used in quasi-experiments. In non-equivalent groups(between-groups) design, participants cannot be part of several groups, i.e., every partici-pant is only subjected to a single treatment. This design lowers the chances of participantsbecome more savvy through practice and experience, which could skew the results.
Being aware of shortcomings of the quasi-experimental design, this sort of studies canbe a very useful tool, especially in situations where randomized experiments are not pos-sible. The design is a very good option to obtain a general overview of the problem andthen follow up with a case study or quantitative experiment to focus on the underlyingreasons for the attained results.
For a modeling exercise, such the one we are dealing with, quasi-experimental de-sign can be applied since the figures and results generated, allow some sort of statisticalanalysis. In addition, without the need of randomization to be undertaken, the designreduces the time and resources for the experimentation.
Quasi-experiments with non-equivalent groups are often used when interventionsare carried out on academic context and the groups correspond to different classes, suchis the case for HA1b hypothesis. The time and resources available, as well as the academicrules constraint the way the groups are formed. So, usually it is not feasible, during thesemester while students are taking the course, to constitute a control group.
In the current modeling exercise, a possible usage of a control group would be anexperiment in which one would expect assessing the benefits of modeling with support ofa BPMN well-formedness rules verification tool. A treatment group would be submittedto an intervention using the BPMN verification tool, while a control group would modelwithout the tool. The ultimate aim of this experiment, would be the assessment andcomparison of generated faults on BPMN models produced by both groups.
In the hypothesis HA1b addressed here students are assigned to non-equivalent groups,all submitted to the same treatment. The division of subjects between groups was conve-nient to cause as little disruption as possible. Albeit the groups’ probabilistic equivalenceis lost, one can still compare the groups. So, to constitute the non-equivalent groups,we only have to assign students based in their graduation. During the experiment, thetwo sets of modelers with different backgrounds, produce BPMN models using the sametesting factor: the proposed exercise. One group, constituted by surrogate of process an-alysts, is more knowledgeable of organizational processes and has more domain skills.The other group, constituted by surrogates of process implementers, has modeling pro-ficiency and previous knowledge in other graphical notations, such as UML, as well asprogramming skills. We would expect that any deviation in results attained by groups,would be due to the fact that BPMN is not equally suitable for process analysts and pro-cess implementers, unlike advocated by the BPMN standard [BPM11].
The assessment of interventions’ effectiveness on BPMN models, from process mod-elers with different skills, is ensured by a post-test. Both groups received the treatment,over the same period of time, and undergone exactly the same support by the teacher
188
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.5. Second Empirical Study
monitoring the experiment. Statistical analysis will determine whether the characteris-tics of the groups had a significant effect in the attained results.
The concise representation of the post-test quasi-experiment with non-equivalentgroups and between-subjects design is the following:
N1 X ON2 X O
8.5.1.5 Collection Procedure
The data for sample S2 is based on BPMN models collected in laboratory classes intendedfor the realization of an exercise on process modeling (Appendix G). The BPMN modelswere produced in a time frame of two hours, by undergraduate students attending twodifferent degrees, in the second semester of the academic year.
The students behave as surrogates of process modelers with different backgrounds.They should complete a provided baseline of the exercise’s solution, by designing newrequired BPMN models, which should be delivered in a file format that could be read bythe Enterprise Architect tool.
After collecting all the repositories with the participants’ solutions, we can now in-stantiate the BPMN models and verify the well-formedness rules violations. The verifi-cation process is made in two ways:
• manually, by visual inspection of delivered BPMN models. This allows the compar-ison with the next method of BPMN models verification, for measurement of theaccuracy of BPMN experts detecting rules violations;• automatically, using a BPMN model checker tool in conjunction with the Framework
for BPMN Model-based testing (see Figure 7.4 in section 7.2.2). This allows theverification of BPMN models against the enriched BPMN metamodel with well-formedness rules as OCL invariants, complemented with best practices from thefield.
The well-formedness rule violations detected in both verifications of the instantiatedBPMN models are stored in a text file, for further processing and transformation intoan SPSS data file. Finally, the statistical treatment required for testing the formulatedhypotheses can be carried out.
8.5.1.6 Analysis Procedure
The techniques chosen for analyzing the experiment are dependent on the adopted ex-periment design (section 8.5.1.4), the variables earlier defined (section 8.5.1.2.2), and theresearch hypotheses to be tested (section 8.5.1.2.1).
189
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.5. Second Empirical Study
8.5.1.7 Instrumentation
The instrumentation process of this empirical study follows closely the one set up for thefirst empirical study. Furthermore, we distributed to the participants training materialregarding BPMN, as well as the modeling exercise detailed in Appendixes F and G
8.5.2 Empirical Study Execution
8.5.2.1 Sample
The sample S2 of BPMN models was collected using students as surrogates for processmodelers practitioners. The experimental work was carried out within the courses onprocess modeling followed by the students. The evaluation of assignment, consisting insolving a modeling exercise, was one of students’ course grades. The students’ participa-tion on the experiment have a didactic objective which was the practical usage of conceptsacquired during the course. All the students that started the experiment concluded it, soany drops occurred before and there were no mortality of subjects. Non-disclosure of theresponses was ensured to the participants.
8.5.2.2 Preparation
Before the conduction of the experimental work on class labs, the students received theproposed problem statement, as well as the Enterprise Architect repository with the base-line of the modeling assignment they were instructed to complete. The students were in-formed about our aim of using, for research purposes, the data collected during the class.However, they were not aware of the aspects being researched, as this could jeopardizethe validity of the results.
8.5.2.3 Data Collection
The students carried out the experiment as part of normal lab class within the course.The exercise’s assignment accounted for their final grade, so there was an incentive toperform it well. The validation of the produced BPMN models did not interfere directlyin the outcome of the solutions, as this was made off-line, after the class. The manualverification of the experiment outputs was carried out by a panel of instructors, based ona previously proposed solution for the problem. We carried out the automatic validationusing the BPMN2USE and USE tools. After the manual and automatic verifications, wasmade available detailed data for the data analysis.
8.5.3 Empirical Study Data Analysis
After data have been collected through samples, we analyzed those samples. This processis detailed in next sections and involves the description of data sets (section 8.5.3.1), con-sideration regarding its reduction (section 8.5.3.2), as well as the testing of the hypotheses
190
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.5. Second Empirical Study
(section 8.5.3.3) defined during the experiments planning. By following these steps, weare instantiating the Analysis Procedure activity, part of the Empirical Study Planning.
8.5.3.1 Data Description
In this section we are going to explore th sample S2 by describing each of the variables inthe data set (Table 8.20).
Table 8.20: Description of variables of the second sample
Variable Description
Modeler Type The BPMN modeler’s background (analyst or implementer)
Total M NOKNumber of distinct BPMN well-formedness rules violationsdetected manually by BPMN experts
Total A NOKNumber of distinct BPMN well-formedness rules violationsdetected automatically by a BPMN model checker
Total M NOK 0.215 51 0.000 0.906 51 0.001Total A NOK 0.143 51 0.011 0.952 51 0.038
a. Lilliefors Significance Correction
8.5.3.1.1 HA1a
In the HA1a hypothesis we want to test whether the effectiveness of BPMN experts ver-ifying BPMN well-formedness rules, has no significant difference from the one providedby automatic model checking.
191
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.5. Second Empirical Study
Our dependent variables are Total M NOK and Total A NOK. Its positive skewnessand kurtosis (Table 8.21) indicates an asymmetric distribution, with a higher frequencyof lower values and a distribution leptokurtic.
Table 8.22 presents the results of two tests which confirm the non-normality of thisvariable. The null hypothesis for each test is based on the assumption that the samplecomes from a Gaussian (normal) distribution. Conversely, the alternative hypothesis isthat the sample comes from a non-normal distribution. With a p-value < 0.05 in both testsfor Total M NOK and Total A NOK variables, we cannot assume the normal distributionof data and therefore we will have to use non-parametric tests to assess hypothesis HA1a.
The Wilcoxon signed-rank test for two paired samples is the non-parametric statistical testthat we use for verifying whether two samples come from the same population. Thesamples that are considered two: (1) well-formedness rules violated according to the ver-ification performed by BPMN experts (variable Total M NOK of the data set); (2) well-formedness rules violated according to the verification executed by an automatic modelchecker (variable Total A NOK of the data set). We will test whether the mean of viola-tions to the BPMN well-formedness rules by each method differ significantly.
8.5.3.1.2 HA1b
In the hypothesis HA1b the dependent variable is Total A NOK and the independent vari-able is Modeler Type. From HA1a hypothesis we already know that the variable is right-skewed and leptokurtic. So, we have also to use non-parametric tests to assess hypothesisHA1b.
The Wilcoxon-Mann-Whitney test for two unpaired samples is the non-parametric statis-tical test that we use for verifying whether the means of two samples differ significantly.The samples considered are the two distinct sets of BPMN models built by surrogates ofprocess analysts and process developers. We will test whether the mean of violations tothe BPMN well-formedness rules in each set differ significantly. In this hypothesis wewant to test whether the quality characteristic of correctness in BPMN models is signifi-cantly different depending on the technical background of process modelers.
8.5.3.2 Data Set Reduction
In the case of the sample S2 it was not performed any data set reduction.
8.5.3.3 Hypotheses Testing
8.5.3.3.1 HA1a
For testing HA1a we have to compare statistics of two paired samples, measure the vari-able in two groups, and test whether the 2 samples mean differs significantly. Thiswill allow us to conclude whether the effectiveness of BPMN experts verifying BPMN
192
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.5. Second Empirical Study
well-formedness rules, have no significant difference from the one provided by automaticmodel checking.
The Wilcoxon signed-rank test for two paired samples is a non-parametric statistical testto test whether the two samples mean differs significantly.
Like the previous test, this one starts by ranking the difference between the observa-tions of the two related samples. Values are ranked from Negative Ranks to Ties. There areno Negative Ranks, which means that the number of violations in BPMN models detectedby manual checking never surpass the automatic checking. Table 8.23 summarizes theinformation concerning the computed ranks.
Table 8.23: Ranks of Total A NOK - Total M NOK for HA1a
a. Total A NOK < Total M NOKb. Total A NOK > Total M NOKc. Total A NOK = Total M NOK
The results for the Wilcoxon signed-rank test for two paired samples we performedare presented in Table 8.24. This test leads us to reject the null hypothesis. On the average,automatic checking is more effective than manual checking unveiling the BPMN modelswell-formedness rule violations. The results of the test are significant, with p-value <0.01.
Table 8.24: Test Statistics of Total A NOK - Total M NOKWilcoxon Signed Ranks Test Sign Test
Z -6.046a -6.784Asymp. Sig. (2-tailed) 0.000 0.00
a. Based on negative ranks.
8.5.3.3.2 HA1b
The hypothesis HA1b concerns whether there are significant differences between the num-ber of BPMN well-formedness rules violations found in process models built by processmodelers with different backgrounds. We have to compare statistics of two unpairedgroups of BPMN models and test whether the two groups mean differs significantly. Ourindependent variable is Modeler Type. We use it to distinguish between the two groups(the BPMN models built by process analysts and the ones built by process implementers).
193
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.5. Second Empirical Study
For testing HA1b we have to compare statistics of two unpaired samples and testwhether the two samples means differ significantly. By doing so, we are testing whetherthe quality characteristic of correctness in BPMN models, has no significant difference,regardless the technical background of process modelers.
To confirm whether the values are statistically significant, we use the Mann-WhitneyU test (Table 8.26). This is a non-parametric test aimed to assess whether two samplescome from the same population. Table 8.25 summarizes the information concerning thecomputed ranks. The test starts by ranking all the observations, disregarding the samplethey come from. The values are sorted in descending order. The columns represent theBPMN models designer (1-Implementer, 2-Analyst), the corresponding count of BPMNmodels (N), their Mean Rank and the sum of their ranks (Sum Ranks).
Table 8.25: Ranks for HA1b
Modeler Type N Mean Rank Sum of Ranks
Total A NOK 1-Implementer 18 27.81 500.52-Analyst 33 25.02 825.5
Total 51
18 of the analyzed BPMN models were built by process implementers, while the re-maining 33 come from process analysts. The Mann-Whitney U test (M-W U) is summa-rized in table 8.26, where in the first column we have the Mann-Whitney U statistic (M-WU), the Wilcoxon W statistic, the test’s Z score, and the 2-tailed asymptotic significance(Asymp.Sig(2-tail)). This test leads us to not reject the null hypothesis.
Table 8.26: Mann-Whitney U test for the Total A NOK variablea
Total A NOK
Mann-Whitney U 264.5Wilcoxon W 825.5Z -0.645Asymp. Sig. (2-tailed) 0.519
a. Grouping Variable: Modeler Type
The test’s results from the Mann-Whitney test are confirmed with a Two-Sample Kolmogorov-Smirnov test. This is a non-parametric test which tests differences in the shapes of thedistributions of the two groups of BPMN models. This test also relies on the rank classifi-cation presented in table 8.25. Table 8.27 presents the test’s results, where we can observethat the most extreme absolute (Absolute) and positive (Positive) difference is 0.177, whilethe most extreme negative difference (Negative) is -0.061. A Kolmogorov-Smirnov Z-scoreof 0.603 and a 2-tailed asymptotic significance (Asymp.Sig.(2-tail)) p-value > 0.05 confirmthe results presented for the Mann-Whitney U test.
These results indicate that the BPMN models produced by process analysts are notsignificantly different regarding well-formedness rule violations than the BPMN models
194
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.5. Second Empirical Study
Table 8.27: Two-Sample Kolmogorov-Smirnov test for the Total A NOK variablea
Total A NOK
Most Extreme Differences Absolute 0.177Positive 0.177
Negative -0.061Kolmogorov-Smirnov Z 0.603Asymp. Sig. (2-tailed) 0.86
a. Grouping Variable: Modeler Type
produced by process implementers.
8.5.4 Empirical Study Results
The discussion regarding the second experimental study is focused on the interpretationof the results (section 8.5.4.1), as well as the limitations of the study (section 8.5.4.2), theinference to be made regarding the extent to which the study’s results are expected tobe representative of the population (section 8.5.4.3), and the identification of the learnedlessons (section 8.5.4.4).
8.5.4.1 Interpretation
Below we go through each of the hypothesis of the second empirical study, by presentingand discussing the outcome of the hypotheses testing.
• RA1a (wrt HA1a) – Empirical evidence allows us to reject the hypothesis that BPMNexperts have the same effectiveness evaluating BPMN rules violations as an auto-matic model checker.We believe this is due to the limitation of process modelers in dealing with largeamount of constructs and well-formedness rules that are part of the BPMN standard(Appendix E). This result is in line with the previous one (RA1), which claims thatBPMN models with many constructs require automatic model checking.• RA1b (wrt HA1b) – Empirical evidence did not allow us to reject the hypothesis that
BPMN models built by process analysts and process developers have the samequality.This is in line with the claimed suitability of BPMN as process modeling amenable,indeed, both for process analysts and process developers [BPM11]. Indeed, wefound process analysts as able to cope with the usage of the BPMN constructs andrules as process implementers. However, since both incur in faults, during pro-cess modeling, an automatic BPMN model checker could drive process modelersthrough the modeling process and give hints when BPMN models are built, allow-ing the achievement of correct models.
195
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.5. Second Empirical Study
8.5.4.2 Validity Threats
In this section we identify the threats and potential weaknesses of the experimental studydescribed using sample S2.
8.5.4.2.1 Internal Validity
Concerning internal validity, we consider two distinct threats: the social threats and themultiple group threats.
Social threats to internal validity can result from usage of distinct treatments withinthe sample, when such distinction produces a behavior’s change in subjects. In S2 noneof the groups was aware of the work performed by the other group so we dismiss any im-itation in groups’ behavior. Moreover, have been made efforts in order that both groupsreceive the same sort of treatment avoiding behaviors of compensatory equalization, com-pensatory rivalry or resentful demoralization from any group.
Multiple groups threats. In S2 we consider the potential of having multiple groupthreats since there are two process modelers’ groups with different backgrounds.∗ Maturation. This threat conveys the distinct reaction from subjects as time progresses,
resulting from, for example, the learning process or saturation. In S2 the risk of het-erogeneous maturation of the process modeling language’s concepts was mitigatedthrough a preliminary phase where modelers got acquainted with BPMN, as well aswith the problem domain (see Appendix F). The preliminary phase lasted for a month,and allowed students to understand the modeling exercise (modeling a solution forthe exercise described in Appendix G) and practice the BPMN language using a designtool. This setting provided homogeneity, in terms of process modeling knowledge, aswell as immersion upon the problem domain, for all the participants in the experiment.∗ Instrumentation. The participants used the same BPMN design tool, thus, problems
with instrumentation could only have been occurred from misusage of the tool. Minordeviations to the instructions provided, regarding the modeling exercise, were cor-rected before verification of BPMN faults.
8.5.4.2.2 External Validity
External validity refers to the researcher’s ability for generalize the results beyond thescope of the experiment. As potential sources of threats are considered:∗ People. In this study, all subjects were students. We believe students can be seen as
surrogates for beginner professional modelers, in the context of this experiment, with-out affecting significantly its outcome. The process modeling role in general is yet toachieve a mature status. So, the problems faced by students do not noticeably differfrom the ones addressed by actual practitioners.∗ Setting. The experiment can be jeopardized when an obsolete experimental environ-
ment is used. In this study, subjects used a case tool environment during modeling,
196
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.5. Second Empirical Study
with limited capabilities concerning syntax verification, in order to challenge theirBPMN knowledge regarding BPMN well-formedness rules and syntax. Another po-tential issue is the usage of a modeling exercise in the experiment, which frames andconsolidates, in a single place, information that otherwise had to be collect and system-atized by process modelers. The provision of a baseline to students can also bias theirsolutions.
8.5.4.2.3 Construct Validity
With respect to construct validity, we consider the social threats.Social threats are the result from the behavior of the subjects and experimenters,
when as consequence of the experiment, they act differently than they would. This be-havioral change, although unintentional, is the result of the subjects’ and experimenters’awareness regarding the experiment. In the case of S2 we are aware of facing the follow-ing threat:∗ Experimenter’s expectancies. The experimenter’s interest in the outcome of the experi-
ment, may bias its execution toward confirming (or refuting) the underlying theory.We did not expect to observe significant differences in the results obtained, since thetraining and directions provided to all participants were similar. So, no interventionon the conduction of the process modeling exercise was made, in order to not interferewith the experiment outcome.
8.5.4.2.4 Conclusion Validity
These are the kind of threats that are inherent to the usage of statistical tests. In the caseof S2 we are aware of facing the following threat:∗ Reliability of treatment implementation. To mitigate this threat, all subjects had similar
conditions for performing their modeling task. The artifacts were distributed to partic-ipants of each group at the same time so they could do their modeling exercise. Aftersolutions have been collected a single person ran the automatic model checker and apanel of experts evaluated manually the outcomes of the exercise.
8.5.4.3 Inferences
The evidences collected throughout the experiment with the second sample, suggest thatthe results obtained by participants, which revealed similar BPMN modeling skills re-gardless the subjects’ technical background, can be extrapolated for undergraduate stu-dents from the two degrees (Informatics / Technology and Industrial Management) ofESTSetúbal. Assuming that these students have basically a similar academic profile tothose of other polytechnics/universities with equivalent degrees, the results could alsobe generalized to those students. However, this assumption should be tested by replicat-ing this experiment in such institutions.
197
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.6. Conclusion
Evidence referred by [Ga08] also suggests that results gathered from students with asimilar profile of the ones of our experiments could be compared to those obtained bynovice professionals. So, one can expect that our results to hold for the population, aswell. Nevertheless, this inference should be supported through replications in profes-sional environments. Extrapolating the observed behavior for seasoned experimentersrequires that other studies must be conducted.
8.5.4.4 Lessons Learned
A large amount of time and effort was required for preparing and conducting the ex-periments with students. After gathering BPMN models we were able to feed them intoour pipeline of applications to support our analysis. We used also transformers for au-tomating the repetitive tasks (e.g. BPMN2USE ans XPDL2USE), namely for generatingthe meta-objects and meta-associations to instantiate the BPMN metamodel.
While conducting the experiment, we realized that some steps on the experimentalprocess could be improved. Also, some challenges could be tackled, in future replicationsof this experiment. We mention some of them:
• narrow the scope of the modeling exercise presented to process modelers partici-pating in the experiment, in order to collate the number of models’ faults and func-tional requirements fulfillment of models;• consider a new version of the exercise, with a treatment that gives to the partic-
ipants free will regarding the model checker tuning. Therefore, modelers coulddecide which rules they wanted to enforce upon BPMN models (e.g. control-flowvs. data flow, or sets of best-practices or standard rules);• measure the effects on students’ learning curve of the BPMN model checker usage.
During the whole experimental process, the practical details of the experimental pro-cess were registered. The feedback provided by students while learning and applyingBPMN to the modeling exercise helped writing this dissertation. These information wereparticularly valuable for packaging the experiments, as well as for replicating the exper-iment in the future.
8.6 Conclusion
Given the BPMN high expressiveness, it is a challenging endeavor a process modeler beable to produce BPMN models conforming well-formedness rules. The process modelerhas to ensure that the BPMN model is both syntactically valid, and that it follows thewell-formedness rules described in BPMN standard.
This chapter was intended to assess, through empirical studies, the effectiveness ofwell-formedness rules formalized in chapter 5, in raising the quality of BPMN models.
In section 8.2 we discussed our option for the scientific method as the research method
198
8. EMPIRICAL STUDIES ON BPMN VERIFICATION 8.6. Conclusion
for validation of the effectiveness of the formalized well-formedness rules. We also pro-posed the BPMN empirical study framework, based in previous studies on ExperimentalSoftware Engineering. Throughout the remainder of the chapter we have instantiatedthe above mentioned framework with two experimental studies, using each one a differ-ent sample (S1 and S2).
For both experimental studies, in the Empirical Studies’ Definition phase, the researchproblem was aligned with their objectives, and the context of the experiments was de-fined (section 8.3). The Empirical Study Planning consisted in a set of activities regard-ing the specification of how the experimental study had to be performed. This wasdone through more detailed decisions concerning the context of the experimental study,namely the formulation of the hypotheses under study, the elicitation of the set of inde-pendent and dependent variables that were used in the statistical tests, the selection ofsubjects to participate in the experiment, the experiments’ design and instrumentation,as well as a preliminary evaluation of the experiment’s validity (sections 8.4.1 and 8.5.1).
The Empirical Study Execution consisted in the instantiation of the previously estab-lished plan, constrained to the specific circumstances found in the actual experiment (sec-tions 8.4.2 and 8.5.2). In the Empirical Study Data Analysis the activities were the data setdescription, its reduction, and the hypotheses test defined during the experiment plan(sections 8.4.3 and 8.5.3). Finally with the Empirical Study Results, the activities were theresults’ packaging so that they could be used in the real world. This involved docu-menting the whole experimental process, and discussing the achieved results, focusingon particular perspectives such as the results’ interpretation, the study’s limitations, theresults’ inferencing to the population, as well as the identification of lessons learned (sec-tions 8.4.4 and 8.5.4).
199
[This page is intentionally blank]
9Empirical Study on BPMN
Measurement
"No one has the right to destroy another person’s belief by demanding empiricalevidence."
Context: Generally process models’ measures are derived, focusing almost exclusivelyon their theoretical or empirical validation. This approach usually leads to unsoundresults difficult to interpret and generalize.Objective: The aim is to empirically validate BPMN measures previously formalizedand theoretically validated, in order to assess whether these measures could be used forprediction of BPMN models’ correctness.
201
9. EMPIRICAL STUDY ON BPMN MEASUREMENT 9.1. Introduction
Method: An empirical study is conducted using the BPMN empirical study framework,in order to validate the adequacy of a set of BPMN base measures to characterize theBPMN models’ internal attributes.Results: The empirical study unveiled the association between the internal attributesof BPMN models and well-formedness rules violations. A predictive model was builtusing a Binary Logistic Regression model for estimating the probability of a BPMN modelhaving errors, based in measures of its internal attributes.Limitations: The validation of results was done, given the restrictions in using actualdata, with a small sample of BPMN models from open repositories.Conclusion: Despite the promising results attained, more research is needed in order tocorroborate the outcomes of the experiment. So, the replication of the studies in industryshould be the next step to ensure the generalization of results.
9.1 Introduction
In chapter 6, we set up BPMN measures focusing mainly on their theoretical validation.The proposal was supported by a framework for BPMN measurement, inspired in themeasure definition process GQM/MEDEA.
Based on knowledge of the BPMN language, specific measurement goals were explic-itly defined and a set of intuitive hypotheses were formulated. A set of base measureswere also identified and validated against the set of theoretically properties. The basemeasures were intended to quantify various BPMN models’ internal attributes (e.g. size,autonomy, etc.) and act as independent variables of a prediction model for BPMN mea-surement. Moreover, an indirect measure for quantifying the dependent variable (well-formedness rules violations) was also derived and related with an external attribute (cor-rectness) of BPMN models.
The results attained in chapter 6 need to be empirically validated. This is the intent ofthe current chapter. So, the main contribution of this chapter is performing a field studyaiming to: (1) the empirical validation of BPMN measures theoretically validated; (2) toformulate a prediction model to explain the relationship between BPMN models’ basemeasures and the models’ external quality measure.
This chapter’s structure follows closely the framework for reporting BPMN experi-ments proposed in section 8.2.1. The empirical study inception, as well as previous re-lated work are discussed in section 9.3. Section 9.4 describes the experimental design,namely the goals (9.4.1) and hypotheses tested (9.4.2). After empirical studies execution(section 9.5) results are presented in section 9.6 and interpreted in section 9.7. Finally,conclusions are drawn in section 9.8.
202
9. EMPIRICAL STUDY ON BPMN MEASUREMENT 9.2. BPMN Measurement Empirical Study
9.2 BPMN Measurement Empirical Study
As mentioned before, this empirical study follows the BPMN empirical study framework.The rationale of each step of the framework was already detailed when empirical studieswere carried out in the previous chapter. To avoid redundancy in the text of this disserta-tion, in the following empirical study we will refer to the sections of the previous chapterwhere was made the foundation of each step of the framework.
We also assume that some of the activities regarding the current empirical study arealready completed, namely the ones related with data collection of the sample. Since wewill use in this chapter the first sample (BPMN models downloaded from open reposi-tories) of the previous chapter, we will also omit the description of the activities alreadydescribed.
9.3 Empirical Study Definition
After a previous work concerning the formalization and implementation of BPMN mea-sures using OCL (chapter 6), the motivation of this chapter is to empirically validatethose derived measures. This will be done through an empirical study on the set of for-malized measures instantiated upon a sample of BPMN process models, collected fromopen repositories, which were already used in chapter 8.
In this section we address successively:• the research problem – justifying the importance of the current empirical study;• the research objectives – highlighting the aims of the study; and• the context definition of the study – defining a framework that constrains the pre-
vious two (research problem and objectives).
9.3.1 Addressing Research Problem
After having previously established the topics of interest of this dissertation (section1.2.1), we focus our study here in a more narrow research topic by tackling only oneof the problems that was then identified (RPB). This will help us in the formulation ofa set of related research hypotheses (section 9.4.2) to be tested (section 9.6.2). So, ourconcerns here are:• associating the internal attributes (i.e. the candidate independent variables) of
BPMN models with the correctness, the external attribute of BPMN models (i.e.the dependent variable).• derive a predictive model for quality characteristic of correctness of BPMN models
based on their internal characteristics.For internal attributes measurement we use the measures formalized and implemented
in chapter 6. For measurement of external attribute of correctness we use the set of BPMNwell-formedness rules previously formalized in chapter 5. The samples used in the em-pirical study are referred in section 7.3.2.
203
9. EMPIRICAL STUDY ON BPMN MEASUREMENT 9.3. Empirical Study Definition
From the attained results we would expect to contribute for improving the BPMNmodeling process, providing the process modelers, methods for measuring BPMN mod-els’ attributes, linked to their proven quality.
9.3.2 Addressing Research Questions and Objectives
In order to address the problems mentioned in the previous section, the following generalquestions are raised:
• Is it possible to establish an association among the process model’s characteristics(e. g. autonomy, complexity, size) and well-formedness rules violations?• Is it possible to define a predictor model for the existence of rule violations in a
BPMN model?
To be more systematic and to rigorously address the above mentioned problems, aswell as to precisely delimit the boundaries of current empirical study, we formulate aset of research questions (RQ). These research questions are in line with the intuitivehypotheses formulated in section 6.4.1.2. The research questions also intend to guide theempirical study through a short descriptive rationale, and detail the research questionRQB, formulated in general terms on chapter 1. More complex research questions (RQBi)were break down in partial questions (RQBij where j is a letter meaning a research sub-question).
• [RQB1]: Is there any association between specific internal attributes of BPMN mod-els and their correctness?∗ [RQB1a]: Is there any association between tangle and correctness of BPMN models?∗ [RQB1b]: Is there any association between autonomy and correctness of BPMN mod-
els?∗ [RQB1c]: Is there any association between complexity and correctness of BPMN
models?∗ [RQB1d]: Is there any association between modularity and correctness of BPMN
models?∗ [RQB1e]: Is there any association between size and correctness of BPMN models?• [RQB2]: Is it possible to produce an accurate prediction model relating internal at-
tributes and correctness of BPMN models?
The remainder of this chapter is dedicated to provide answers for the research ques-tions introduced in this section. The latter will be refined into research goals (section9.4.1) which, in turn, will lead to the specification of the research hypotheses (section9.4.2).
To provide grounded answers to the aforementioned research questions, we are goingto analyze the sample of BPMN models detailed in Appendix D, collected from disparatesources (section 7.3.2). For the experiments’ definition we apply the GQM framework[BCR94], introduced in section 6.3.2.
204
9. EMPIRICAL STUDY ON BPMN MEASUREMENT 9.4. Empirical Study Planning
The instantiation of the GQM framework for the empirical study of this chapter, isbased in the research objective GB, as the GQM’s top-level goal (section 6.4.1.1). The goalis stated according to the terms of the GQM template. For the sake of readability, wereplicate here the GQM top-level goal already defined:
Analyze BPMN modelsfor the purpose of predictionwith respect to assessment of quality characteristic of correctness,from the point of view of process modelers,in the context of a sample from open repositoriesconstrained by the requirements of regression studies.
9.3.3 Context Definition
We follow the principles stated in section 8.3.3 and use the first sample mentioned thereand identified as S1.
9.4 Empirical Study Planning
This section is concerned with how the experiment will be performed. Before describinghow the experiment was conducted (section 9.5), we will detail the environment in whichwe will pursuit the experiment.
This will be the basis for introducing, in the following sections, the research goals(section 9.4.1), the hypotheses under study and the set of independent and dependentvariables to be used for hypotheses’ evaluation (section 9.4.2), the criteria for selectionof subjects participating in the experiment (section 9.4.3), the experiments’ design andinstrumentation (section 9.4.4), and an evaluation of the experiment’s validity (section9.4.6).
The outcome of this section is an experimental design, i.e., a receipt for the exper-iment, which provides the information for experiment’s replication, and also to allowreaders to evaluate the experiment’s internal validity.
9.4.1 Goals
The research objective GB outlined in section 9.3.2 is refined here as research goals. Theseresearch goals, are the subgoals of GB1 (decomposed as GB1a – GB1e) and GB2 (section6.4.1.1).• GB1 (wrt RQB1) – analyze BPMN models for the purpose of measurement of mod-
els’ internal attributes with respect to assessment of quality characteristic of cor-rectness, from the point of view of process modelers, in the context of samples ofmodels collected from open repositories constrained by the requirements of regres-sion studies.Question: How to assess the internal attributes of BPMN models?
205
9. EMPIRICAL STUDY ON BPMN MEASUREMENT 9.4. Empirical Study Planning
∗ GB1a (wrt RQB1a) – . . . for the purpose of measurement of models’ tangle . . .Question: How to measure the tangle of BPMN models?
∗ GB1b (wrt RQB1b) – . . . for the purpose of measurement of models’ autonomy . . .Question: How to measure the autonomy of BPMN models?
∗ GB1c (wrt RQB1c) – . . . for the purpose of measurement of models’ complexity . . .Question: How to measure the complexity of BPMN models?
∗ GB1d (wrt RQB1d) – . . . for the purpose of measurement of models’ modularity . . .Question: How to measure the modularity of BPMN models?
∗ GB1e (wrt RQB1e) – . . . for the purpose of measurement of models’ size . . .Question: How to measure the size of BPMN models?
• GB2 (wrt RQB2) – analyze BPMN models for the purpose of producing a predic-tion model with respect to assessment of the quality characteristic of correctness,from the point of view of process modelers, in the context of samples of mod-els collected from open repositories, constrained by the requirements of regressionstudies.Question: How to evaluate a predictive model of BPMN models’ quality character-istic of correctness?
9.4.2 Hypotheses and Variables
The goals settled in section 9.4.1 guide us through the derivation of six different hypothe-ses. For each of them, we formulate the null hypothesis (denoted by the index 0), leavingthe alternative hypothesis (index 1) to be inferred by negating the null one.
• HB10 (wrt GB1) – the external attribute of BPMN models, correctness, has no signifi-cant correlation with BPMN models’ internal attributes.∗ HB1a0 (wrt GB1a) – BPMN models’ correctness has no significant correlation with
BPMN models’ tangle.∗ HB1b0 (wrt GB1b) – BPMN models’ correctness has no significant correlation with
BPMN models’ autonomy.∗ HB1c0 (wrt GB1c) – BPMN models’ correctness has no significant correlation with
BPMN models’ complexity.∗ HB1d0 (wrt GB1d) – BPMN models’ correctness has no significant correlation with
BPMN models’ modularity.∗ HB1e0 (wrt GB1e) – BPMN models’ correctness has no significant correlation with
BPMN models’ size.• HB20 (wrt GB2) – The predictive model of BPMN models’ correctness has no signifi-
cant difference from the actual correctness of BPMN models.
For selecting the variables related with the formulated hypotheses, we rely also inthe GQM framework [BCR94], namely on the specified research goals. The dependentvariables’ elicitation is guided by the quality focus of each goal, which is preceded by thewith respect to sentence in each research goal’s formulation.
206
9. EMPIRICAL STUDY ON BPMN MEASUREMENT 9.4. Empirical Study Planning
The independent and dependent variables used for each identified hypothesis are de-tailed in the table 9.1. Each dependent variable is specified in terms of the constructs ofthe BPMN metamodel and formally defined through an OCL expression which allowsits computation. We refer to the definition of each variable, the attributes to be mea-sured, the measurement to be made, the computation rule to be applied, and the unit ofmeasurement assigned (see the description column in table 9.1).
Table 9.1: Variables for HB1a to HB1eRole Variable Name Description
Indep. TangleA ratio variable with properties specified in theDefinition 6.7 and measured according Defini-tion 6.12.
Indep. AutonomyA ratio variable with properties specified in theDefinition 6.8 and measured according Defini-tion 6.13.
Indep. ComplexityA ratio variable with properties specified in theDefinition 6.9 and measured according Defini-tion 6.14.
Indep. ModularityA ratio variable with properties specified in theDefinition 6.10 and measured according Defini-tion 6.15.
Indep. SizeA ratio variable with properties specified in theDefinition 6.11 and measured according Defini-tion 6.16.
Dep. Total S NOK
A ratio variable that conveys the correctnessin the BPMN model, according to the verifica-tion made by an automatic model checker. Thewell-formedness rules are formulated in termsof OCL invariants (see Appendix E for the com-plete list of rules)
9.4.3 Subjects selection
For subjects selection in this experimental study we rely on the sample S1 collected forthe study described in section 8.4.1.3.
9.4.4 Experimental Design
As referred in section 8.4.1.4, the hypotheses and variables we elicited in section 9.4.2place restrictions on the experiment design to choose. On the other hand, this constrainsthe statistical studies to pursue the analyze the data collected from the experiment.
Given the sample description and the nature of the phenomena in study (correctnessof BPMN models), we apply a correlational design in this study to collect and analyzedata in search for associations between variables. The intention of the study is to show,whether variables representing internal attributes (e.g. complexity, size) and externalattributes (correctness) of BPMN models, have any kind of relationship among them.
A convenience sample, of more than 40 BPMN process models from open repositories,described in section 8.4.1.3, is processed through the use of OCL measures automatically
207
9. EMPIRICAL STUDY ON BPMN MEASUREMENT 9.4. Empirical Study Planning
collected from the instantiated BPMN models.
For HB1a to HB1e hypotheses, we use the convenience sample as a single group. The ar-rangement means that there is no control group in this post-test only non-experimentaldesign. In practical terms we pack all the BPMN models collected from the repositories,apply the measurement of internal attributes, as well as check the correctness of BPMNmodels and verify whether the association between variables is statistically significant.It consists in a one-shot survey with a single observation, which is the simplest form ofnon-experiment. This is the design [Tro06] most adequate for the descriptive study toaddress our goal GB1 and the corresponding research question RQB1.
The concise representation of the non-experiment is the following:
X O
We should conduct further research, regarding other kinds of formulation of pro-cesses, to check which conclusions are specific to the BPMN process models and whichare generalizable to other notations.
9.4.5 Collection Procedure
The motivation for this procedure was described in section 8.4.1.5, where it was alsopresented an overview of the procedure followed for collecting the sample S1 used inthis experiment.
9.4.6 Analysis Procedure
The analysis techniques chosen for the current experiment depend on the adopted ex-periment design (section 9.4.4), the variables defined and the research hypotheses beingtested (section 9.4.2).
The statistical tests and their conditions for the current experiment are the same asthose defined in section 8.4.1.6.
9.4.7 Instrumentation
The instrumentation process is concerned with the artifacts that are used during the cur-rent experiment. It includes, among other, off-the-shelf and custom made tools that sup-port measurements in the experiment. It also includes the logistics required to put inplace the replication of the experiment. These instruments are described in section 8.4.1.7.
The process models in Figures 7.8 and 7.11 detail the role of each mentioned tool, usedin a pipes and filters architectural style [BCK03].
208
9. EMPIRICAL STUDY ON BPMN MEASUREMENT 9.5. Empirical Study Execution
9.5 Empirical Study Execution
The Empirical Study Planning instantiation for the current experiment was synchronizedwith the instantiation of the plan regarding the experiments of the previous chapter (de-scribed in section 8.4.2). So, all the steps followed (Sample, Preparation, and Data Collec-tion) were the same for the experiments reported on both chapters.
9.6 Empirical Study Data Analysis
After we have collected the sample, we analyzed it. This process is detailed in the nextsections and involves the description of the data set (section 9.6.1), as well as the testingof the hypotheses (section 9.6.2) defined during the experiment’s planning. By follow-ing these steps, we are instantiating the Analysis Procedure activity, which is part of theEmpirical Study Planning (section 9.4).
9.6.1 Data Description
Data description helps understanding the sample gathered. A detailed data descriptionof the variables collected in our sample is presented in Table 9.2.
As mentioned before, relevant descriptive statistics are the measures of central ten-dency, as well as dispersion measures and data distributions like normal distribution.We said in section 8.4.3.1 that a sampling distribution is not normally distributed if it isskewed or has outliers which influences the statistical tests to choose in the subsequentanalysis.
We begin exploring the current sample by describing in Table 9.2 each of the relevantvariables in the data set.
Table 9.2: Description of variables of measures’ sampleVariable Description
Source The company that was the source of the model
ModelBPMN model name, which indicates the domain of theBPMN model
Total S NOKNumber of distinct BPMN well-formedness rules violationsdetected in the model by the BPMN automatic checker
Tangle Computed according equation 6.1Autonomy Computed according equation 6.2Complexity Computed according equation 6.3Modularity Computed according equation 6.4Size Computed according equation 6.5
isFaultyModelObtained by transformation of variable Total S NOK. is-FaultyModel was set to 0 (Error Free) for Total S NOK equal0 and, to 1 (Error Found) for Total S NOK greater than 0
Looking for differences concerning measures of BPMN models from distinct sources,one can see in Figure 9.1 that, on average, models from Bizagi reveal higher values of the
209
9. EMPIRICAL STUDY ON BPMN MEASUREMENT 9.6. Empirical Study Data Analysis
Complexity measure, equal Size measure value and lesser values of the other measures,when compared with models from Trisotech.
Figure 9.1: Radar diagram depicting the BPMN measures by Source
In Table 9.3 we summarized the descriptive statistics of the independent variablesconsidered for testing the hypotheses related with this sample. The normality tests ofthose variables are also presented (Table 9.4). Next we discuss the statistics for the hy-potheses.
a. Multiple modes exist. The smallest value is shown
In the HB1a to HB1e hypotheses we want to test whether the BPMN models correctnessis associated with certain internal measures of BPMN models. The dependent variablefor these hypotheses is Total S NOK, which was already analyzed in section 8.4.3.1.1.
210
9. EMPIRICAL STUDY ON BPMN MEASUREMENT 9.6. Empirical Study Data Analysis
Its positive skewness (value 1.746 on Table 8.5) indicates an asymmetric distribution,with a higher frequency of lower values. The value of the kurtosis (value 3.19 on Table8.5), allows us to say that the distribution is also leptokurtic, with higher probability ofvalues being close to the mean than in a normal distribution. Both the distribution’sskewness and kurtosis provide a clue on the non-normality of the variable’s data. Table8.6 presented also in section 8.4.3.1.1 confirms the non-normality of the Total S NOKvariable.
Besides the dependent variable Total S NOK, we have also the independent vari-ables Tangle, Autonomy, Complexity, Modularity, and Size. The positive right-skewed andleptokurtic of Autonomy, Complexity, Modularity indicates the non-normality of this vari-able (values of skewness/kurtosis in Table 9.3). Table 9.4 presents the results of normalitytests that confirms its significance and thus, we cannot assume a normal distribution ofAutonomy, Complexity, and Modularity. Only Tangle and Size conform with a Gaussiandistribution.
The non-parametric tests of Spearman’s rank correlation coefficient is used to assess HB1a
to HB1e hypotheses. Spearman’s coefficient assesses how well the relationship betweentwo variables can be described using a monotonic function. We will test whether exists acorrelation between the internal attributes of BPMN models (through the measures Tan-gle, Autonomy, Complexity, Modularity, and Size) and the external attributes of BPMN mod-els (correctness) measured through the counting of violations of BPMN well-formednessrules.
Table 9.4: Tests of Normality (III)Kolmogorov-Smirnovaa Shapiro-Wilk
a. Lilliefors Significance Correction*. This is a lower bound of the true significance.
9.6.2 Hypotheses Testing
9.6.2.1 Hypotheses HB1a to HB1e
For testing HB1a to HB1e, we should find whether the variable representing the external at-tribute correctness in BPMN models (Total S NOK) is associated with the variables thatrepresent the internal attributes of BPMN models (Tangle, Autonomy, Complexity, Modu-larity, and Size).
We perform the correlation analysis, using the Spearman’s correlation test. For each
211
9. EMPIRICAL STUDY ON BPMN MEASUREMENT 9.6. Empirical Study Data Analysis
of the hypotheses considered, we computed and tested the correlations of the depen-dent variable (Total S NOK) with respect to the independent variables, as summarizedin table 9.5. We realized significant correlations between the independent variable and thedependent variables Tangle, Autonomy, Complexity, and Size. Only for Modularity there isno significant correlations with Total S NOK.
N 48 48 48 48 48Total S NOK Correlation Coefficient .359* -.345* .509** 0.053 .485**
Sig. (2-tailed) 0.012 0.016 0.000 0.723 0.000
**. Correlation is significant at the 0.01 level (2-tailed).*. Correlation is significant at the 0.05 level (2-tailed).
We can conclude that the internal attributes of BPMN models Tangle, Autonomy, Com-plexity, and Size, are associated to the external attribute correctness of BPMN models.So, one can reject the hypothesis that in BPMN models’ sample, the referred internal at-tributes have not significant correlation with the defects found in the BPMN models, atthe 0.05 level.
9.6.2.2 Predictive Model
In this section we derive a predictive model to statistically detect faults in BPMN models,using a Binary Logistic Regression (BLR) model.
BLR requires an initial calibration phase where the values of the model coefficientsare determined based on preexisting values. For calibration of our model, we use thesample of BPMN models collected from open repositories. These models constituted thetraining set of the statistical model.
Logistic regression (aka logistic model or logit model) is used for prediction of theprobability of occurrence of an event by fitting data to a logistic curve [Fie09]. Here, theconsidered event is a BPMN model containing well-formedness rules violations. BLR is ageneralized linear model used for binomial regression. The logistic function is useful be-cause it can take as an input any value from negative infinity to positive infinity, whereasthe output is confined to values between 0 and 1. BLR has been used in experimentalworks in several domains, such as Software Engineering (e.g. [BAM10]).
After the BLR model has been calibrated, and like other types of regression models,it is fed with a set of numerical or categorical regressors variables (predictor variables).In our case, the regressors are a set of internal attributes of BPMN models (Tangle, Auton-omy, Complexity, and Size) derived in chapter 6, and shown in the previous section to beassociated with the external attribute correctness of BPMN models.
As dependent variable, in the formulation of our predictive model, we define isFault-yModel. This variable represents the probability of a given BPMN model to be correctbased on its own internal attributes. The BLR is formalized by the following expression
212
9. EMPIRICAL STUDY ON BPMN MEASUREMENT 9.6. Empirical Study Data Analysis
The variable z represents the exposure to some set of risk factors (the internal at-tributes of BPMN models in this case), while f (z) represents the probability of a par-ticular outcome (incorrectness of BPMN model due to non-compliance with some well-formedness rules), given that set of risk factors. Variable z is a measure of the total con-tribution of all the risk factors used in the model and is known as the logit.
The β0 parameter is called the intercept parameter (value of z when all risk factors are0) and β1 through β4 are the regression coefficients of the corresponding predictors. Eachof the regression coefficients quantifies the contribution of the respective predictor. Apositive regression coefficient means that the risk factor increases the probability of theoutcome, while a negative regression coefficient means that the risk factor decreases theprobability of that outcome. The values of all βi parameters are obtained by regressionupon the mentioned training set of predictor and outcome values.
The values of the predictors were obtained by running the measures upon the col-lected sample of BPMN models from open repositories. The values of the outcome vari-able (values of isFaultyModel for each BPMN model in the sample set) were obtained bytransformation of the values of variable Total S NOK (Table 9.2). isFaultyModel was setto 0 (Error Free) for Total S NOK equal 0, and to 1 (Error Found) for Total S NOK greaterthan 0.
We are interested in finding high correlations between the internal attributes of BPMNmodels and the external attribute represented by isFaultyModel. Furthermore, we want toensure there is independence among of the base measures of the internal attributes. Todo so, and since, as we saw in Table 9.4, none of the variables has a normal distribution,we use the non-parametric Spearman’s rho correlation coefficient (Table 9.6). We concludethat all correlations are significant, except the relationship between Tangle and Autonomy.
The results presented in Table 9.6 show that all other measures have a low to moderatecorrelation with isFaultyModel. Tangle and Autonomy exhibit low correlation and thereforeseems to be the worst predictors among the chosen measure set.
Noteworthy, the high correlation between Size and Tangle / Autonomy. This is a poten-tial threat since it indicates the presence of multicollinearity, which means that predictorvariables are highly correlated among them.
The collinearity statistics in Table 9.7, allow us to consider all regressors as acceptable.The presence of multicollinearity is only assumed for the thresholds of Tolerance < 0.20
213
9. EMPIRICAL STUDY ON BPMN MEASUREMENT 9.6. Empirical Study Data Analysis
**. Correlation is significant at the 0.01 level (2-tailed).*. Correlation is significant at the 0.05 level (2-tailed).
or VIF > 5.
Table 9.7: Collinearity Statistics for BLR modelTolerance VIF
Coefficients for Dependent Variable: Tangle
Autonomy 0.38 2.634Complexity 0.22 4.544
Size 0.357 2.803
Coefficients for Dependent Variable: Autonomy
Complexity 0.224 4.466Size 0.429 2.332
Tangle 0.374 2.673
Coefficients for Dependent Variable: Complexity
Size 0.946 1.057Tangle 0.445 2.246
Autonomy 0.46 2.175
Coefficients for Dependent Variable: Size
Tangle 0.367 2.727Autonomy 0.447 2.236
Complexity 0.481 2.081
Running BLR, the step 0 of model construction, give us the intercept only model, amodel with no predictors’ variables included (Tables 9.8 – 9.10). The value of 54.2%(= 26/48) in Table 9.8, gives the classification accuracy we get if we had considerer all theBPMN models as Error Free, which is the same as the probability of getting a error freemodel from the sample.
In Table 9.9 the value of -0.167 in the exponentiate intercept, is the intercept B of themodel, since there are no predictor variables. The level of significance of 0.564 for theWald statistics allows not reject the null hypothesis, which states an equal probability offind errors in the model or getting a error free model. The exponentiate intercept, Exp(B),is 0.846 (= 22/26), which means there is a likelihood of 15.4% of not having errors in themodel. Another way to convey the same is saying that process modelers are likely to geta model free of errors at an odds rate of 1.18 (= 26/22).
214
9. EMPIRICAL STUDY ON BPMN MEASUREMENT 9.6. Empirical Study Data Analysis
Table 9.8: Step 0 - Classification Tablea,b
Predicted
isFaultyModel
ObservedErrorFree
ErrorFound
PercentageCorrect
isFaultyModel Error Free 26 0 100Error Found 22 0 0
Overall Percentage 54.2
a. Constant is included in the model.b. The cut value is .500
Table 9.9: Step 0 - Variables in the EquationB S.E. Wald df Sig. Exp(B)
Constant -0.167 0.29 0.333 1 0.564 0.846
According to Table 9.10, one can conclude that all regressors are statistically signifi-cant and related to isFaultyModel.
Table 9.10: Step 0 - Variables not in the EquationScore df Sig.
In Table 9.11 are presented the results of the tested hypothesis that have the the samepredictive capacity of the BLR model when all three independent variables are entered.The null hypothesis for the Omnibus test is that adding the predictors to the model has notsignificantly increased our ability to predict the correctness of the BPMN model. Giventhe statistically significance of the Chi-square statistic, we can reject the null hypothesisand conclude that the predictors have significantly increased our ability to predict thecorrectness of the BPMN model.
Table 9.12, shows the R Squares presented in both tests. These statistics give a roughestimate of the variance in isFaultyModel that can be predicted through the conjunctionof the three variables. The Nagelkerke R Square statistic, which is scaled from 0 to 1.0, canroughly be interpreted as 34.5% of variability on dependent variable be accountable tothe independent variables. The Cox and Snell test is usually an underestimate. The -2 Loglikelihood statistic is quite small, meaning that the model has some predictive capacity ofisFaultyModel occurrence.
The Hosmer-Lemeshow goodness-of-fit test considers as null hypothesis that there is alinear relationship between the predictor variables and the log odds of the criterion vari-ables. Given the level of significance shown in Table 9.13 we reject this hypothesis. Sincethe Hosmer-Lemeshow is not statistically significant, it means that our BLR model has
215
9. EMPIRICAL STUDY ON BPMN MEASUREMENT 9.6. Empirical Study Data Analysis
Table 9.11: Step 1 - Omnibus Tests of Model CoefficientsChi-square df Sig.
Step 15.505 4 0.004Block 15.505 4 0.004
Model 15.505 4 0.004
Table 9.12: Step 1 - Model Summary-2 Log
likelihoodCox & Snell
R SquareNagelkerke
R Square
50.704a 0.276 0.369
a. Estimation terminated at iteration number 5.
some predictive capacity.
Table 9.13: Step 1 - Hosmer and Lemeshow TestChi-square df Sig.
3.064 8 0.93
If we compare the results of Table 9.14 with the ones previously obtained in Table 9.8,we can conclude that the introduction of predictors variables in the BLR model, increasedthe errors’ predictive capacity of the model, from 0% to 72.7%. The Overall Percentage ofpredictive capacity of the BLR also increased by 20.8% (from 54.2% to 75%).
According to Table 9.14, we can say that 76.9% of the BPMN models which are ErrorFree are predicted correctly with this model, while in 72.7% classified as Error Found werepredicted correctly. In other words, we have 6 false positives and 6 false negatives. Wepredict 22 (=16+6) BPMN models with errors, but we fail in 6 cases, so our false positiverate is 27.2% (=6/22). We predict 26 (=20+6) BPMN models without well-formednessrules violations and we fail in 6 cases. Thus, our false negatives rate is 27.2% (=6/26).
We can conclude from Table 9.15 that, when all variables (Tangle, Autonomy, Com-plexity, and Size) are considered together with an α=0.05, they are not significant. Thissuggests some correlation among variables, as we previously saw in Table 9.6. The Waldstatistic, which tests the unique contribution of each predictor in the context of the otherpredictors, shows that Size is the predictor that most contributes to the BLR model, fol-lowed by Autonomy.
The z value required to calculate the probability of a BPMN model being incorrectlyformed (isFaultyModel), is so the one we can obtain by instantiating the equation 9.2 withthe values of the B column of Table 9.15, as follows:
a. Variable(s) entered on step 1: Tangle, Autonomy, Complexity, Size.
recognition, which are based on the understanding and measure of relevance. In this par-ticular case precision (aka positive predictive value) is the percentage of errors detectedby the BLR model that are correct: 72.7% (=16/(16+6)). On the other hand, recall (akasensitivity) is defined as the percentage of errors that are selected, which also takes thevalue of 72.7% (=16/(16+6)).
Since we have a high recall, it means that the BLR model returned most of the relevantresults. The high precision means also that the BLR model returned substantially morerelevant results than irrelevant. In terms of accuracy the BLR model gets the value of 75%(=(16+20)/(16+20+6+6)), which is considered good enough for this kind of application.
9.7 Empirical Study Results
In this section we package the results so they can be used by the process modeling com-munity, as mentioned in section 8.4.4.
We focus the following discussion on the interpretation of the results (section 9.7.1),the inference regarding the extent of the study’s results being expected to hold for thepopulation (section 9.7.2), and the identification of the learned lessons (section 9.7.3).
9.7.1 Interpretation
This section analyzes the outcome of the tests, anchored on the theory formulated inchapter 6. In the following paragraphs we go through the hypotheses to discuss theresults of their testing.
• RB1 (wrt HB1) – Empirical evidence allows us to reject the hypothesis that BPMN
217
9. EMPIRICAL STUDY ON BPMN MEASUREMENT 9.7. Empirical Study Results
models’ external attribute of correctness has no association with the internal at-tributes of Tangle, Autonomy, Complexity, and Size. On the other hand we do notreject the null hypothesis for the measure of ModularityThe relationships are positive (negative) between the BPMN models’ measures ofTangle, Complexity, and Size (Autonomy), and the number of BPMN well-formednessrules violations.The correlation strength of these relationships can be considered low (between 20%–40%), for Tangle and Autonomy, and moderate (between 40%–70%), for internal at-tributes’ of Complexity and Size (Table 9.5).• RB2 (wrt HB2) – A predictive model based in the Binary Logistic Regression, was
built to detect faults in BPMN models.We have shown that using a set of BPMN measures, with a moderate sized trainingset, we were able to build a model that predicted correctly around 72.7% of themethods which are faulty BPMN models and 76.9% of the BPMN models whichare not, with a false positive rate of 27.2% and a false negative rate also of 27.2%.
The potential validity threats of this experimental work were previously describedin section 8.4.4.2, regarding the first sample (S1) used in chapter 8, the same used in thecurrent experiment. Furthermore, albeit BLR is a mathematically sound approach, sincewe used a limited number of BPMN models, we are aware that our ability to generalizethe results is limited.
9.7.2 Inferences
In section 8.4.4.3, we characterized the aim of this activity, which is mainly to deriveconclusions from the statistical tests results and generalize the research.
We expect the same patterns of BPMN models’ internal attributes and correctnessshown through this empirical study should hold and be observable with other modelspublicly available. The results of this study should only be considered valid in the contextof BPMN models, rather than generic to other process notations or formalizations.
It seems also plausible that these observations could also be applied to actual BPMNmodels of real organizations. However, we must bear in mind that observing these resultsin an organization, requires some stability of the modeling tool, as well the modeling pro-cess. The introduction of new tools and working processes could easily alter structurallythe prediction model, forcing the re-computation of the BLR coefficients.
Nevertheless, it was demonstrated in this chapter that it is possible to produce a BLRmodel for estimating the probability of a BPMN model having errors, based in measuresof its internal attributes. However, for applying the BLR model in an organization, itis required the calibration of the BLR coefficients based in the repository of past BPMNmodeling projects.
218
9. EMPIRICAL STUDY ON BPMN MEASUREMENT 9.8. Conclusion
9.7.3 Lessons Learned
In section 8.4.4.4, we already discussed the learned lessons with respect to the opera-tionalization of the experimental studies. Those remarks also applies to this regressionstudy.
Specifically we highlight the use of statistical techniques to calibrate a predictivemodel amenable to determine the probability of a BPMN model to violate well-formednessrules. Moreover, the BLR model besides being able to predict how many faulty modelsexist in a sample, it can also predict the number of false positives and false negatives.
9.8 Conclusion
One of the objectives of this chapter was to perform a correlational field study in orderto complete the theoretical study of chapter 6 regarding BPMN measures, by validatingempirically those measures and the relationship between BPMN models’ base measuresand the BPMN models’ correctness. Another objective was to formulate a predictionmodel for the BPMN models’ correctness based on BPMN models’ base measures.
For achieving the BLR model, was used a sample of BPMN models as a training set,for calibration purposes. Using the measures theoretically validated in chapter 6, theregressors (BPMN measures) were computed, as well as the BLR coefficients needed.Using the probabilistic model, we were able to estimate the probability of a given BPMNmodel being faulty, giving as input the measures of its internal attributes.
In this chapter, to conduct the empirical study, we also followed the BPMN empiricalstudy framework, derived in the previous chapter. In section Empirical Study Definition, theresearch problem was stated with the objectives of the experiment, as well as the contextin which the experiment was carried out. The Empirical Study Planning detailed the deci-sions concerning the context of the experiment, namely the formulation of the hypothesesunder study and the elicitation of the set of independent and dependent variables thatwere used in the statistical tests.
In section 9.6, we described the data set and performed the testing of the hypothesesdefined during the experiment plan. Finally, in section 9.7, after the experiment has beenperformed, the results were packaged so that they could be disseminated.
219
[This page is intentionally blank]
10Conclusion
"I never see what has been done; I only see what remains to be done."
Context: We have concluded the dissertation on quality of process modeling usingBPMN by following a product-oriented perspective and using a model-driven approach.Objective: To survey the work done in the dissertation and forecast future directionsfor that work.Method: We highlight the main aspects covered by the previous chapters, our maincontributions, as well as the future work that we intend to undertake to pursue theopen research path.Results: The list of main contributions and the plan for future research are the mainoutputs of this chapter.Limitations: Being a summary, this chapter only shallowly covers the work done.Conclusion: The chapter gives a glimpse on the work done and to be pursued in searchof quality of process modeling using BPMN.
221
10. CONCLUSION 10.1. Synthesis
10.1 Synthesis
BPMN is the most popular process modeling language. However, it reveals limitations.We had to confirm whether those limitations were relevant enough to affect the quality ofprocess models produced using actual BPMN tools. This was done by checking whethercurrent BPMN tools implementing the BPMN standard could rigorously verify processmodels, and ultimately produce good quality artifacts.
The achieved results corroborate the idea that the limitations in the BPMN standard,i.e. rules informally specified in natural language, contribute for poor quality of processmodels, given the weakness of available BPMN tools in terms of models’ verification.
So, the main contribution intended by this dissertation was the improving of the qual-ity of process models at the design phase.
We add to the BPMN 2.0 metamodel, for enforcing the process diagrams’ quality(namely regarding their correctness): (1) well-formedness rules defined in the BPMNstandard document in natural language; (2) best-practices rules promoted by BPMN ex-perts in the literature.
We also derived and add to the BPMN metamodel measures that could help processmodelers be knowledgeable of BPMN models’ internal attributes (e.g. complexity, size).These measures can also provide process modelers with hints and guidelines for improv-ing models’ external quality characteristics (e.g. correctness).
Grounded in the research problems and research questions, the purpose of the dis-sertation was summarized in the following thesis statement: in order to improve BPMNmodels’ quality we propose a model-driven approach capable of formalizing: (i) well-formedness rules; (ii) measures for assessment of models characteristics (chapter 1).
In chapter 2 current process modeling languages were surveyed and assessed. Fromthose languages we chose BPMN to assess the quality of process models. However,BPMN has weaknesses that we confirmed through a survey upon BPMN tools (chapter3).
In chapter 4 we conducted a systematic review regarding BPMN process models ver-ification. We concluded that most of the research works surveyed addressed essentiallythe properties verification aspects, i.e., the papers were related with checking of domain-independent properties (e.g. deadlock, liveness). The aspects considered relevant forthis dissertation were insufficiently covered by those studies, namely the BPMN well-formedness/best-practices rules checking, as well as the measurement dimension, andempirical validation of BPMN process models.
Another literature review was made regarding the research work done about mea-surement of BPMN models quality characteristics. The limitations pointed out to thesurveyed work were: (1) no theoretical validation of proposed measures; as well as (2)the computation of measures without considering the well-formedness of BPMN mod-els. To circumvent these limitations, it was proposed the elicitation of a set of BPMNmeasures aligned with the BPMN 2.0 standard.
222
10. CONCLUSION 10.2. Contributions
In chapter 5 we formalized and implemented BPMN rules, informally specified in theBPMN 2.0 standard, as well as best-practices rules, advocated by BPMN practitioners.Since property verification (e.g. deadlock, liveness) is a topic already covered by severalother approaches, it was not addressed by the present dissertation.
In chapter 6 we theoretically validated a set of BPMN measures through the BPMNmeasurement framework. Following the guidelines of the framework, we defined mea-surement goals and a set of intuitive hypotheses to be validated. Next, we also identifiedinternal attributes of interest and validated related base measures against a set of theoret-ically properties. The base measures were intended to quantify BPMN models’ internalattributes. They were also designated as independent variables of a prediction modelfor BPMN models’ correctness. Finally, we derived indirect measures for quantifying thedependent variable related to an external attribute (correctness) of BPMN models. Theproposed framework for BPMN models measurement requires further instantiation withother measures, in order to assess its general suitability.
Chapter 7 brought BPMN to the context of MDE, and presented a framework forBPMN model-based testing. The MDE approach was instantiated through BPMN well-formedness rules implementation, as well as with the definition and construction of thetransformations needed for data collection to allow empirical studies in the followingchapters. Albeit the approach brought the BPMN to the context of MDE, using differentkinds of tools and languages, we are aware that more work can be done regarding processmodeling transformations, using DSLs, and their formalization.
Chapter 8 presented empirical evidence of the feasibility of a MDE approach to BPMNmodel checking. The well-formedness rules formalized in chapter 5 were used for thatpurpose. The overall conclusion is that well-formedness rules have a significant impactupon the final quality characteristic of BPMN models’ correctness.
In chapter 9 was performed a regression analysis in order to provide empirical evi-dence of the usefulness of the BPMN measures proposed in chapter 6. Those measureswere used as exploratory variables in a Binary Logistic Regression model that allows toforecast potentially defective BPMN models.
The samples used in the experiments of the two previously mentioned chapters werecollected from examples hosted in repositories from tool makers, as well as BPMN mod-els built by students, as surrogates of professional BPMN process modelers. Furtherreplication studies must be done to corroborate the achieved results.
10.2 Contributions
In the following sections we summarize some of the major and minor contributions of ourdissertation. The classification assigned has essentially to do with the expected impactand future reuse of the contribution, rather than the work required to achieve it.
223
10. CONCLUSION 10.2. Contributions
10.2.1 Major Contributions
10.2.1.1 BPMN Rules Formalization
A formalization of BPMN well-formedness rules as OCL invariants, was presented insection 5.2. The derived well-formedness and best-practices rules were attached to theBPMN metamodel in order to enhance the BPMN models’ verification.
10.2.1.2 BPMN Measures
The base measures are intend to quantify various BPMN models’ internal attributes (e.g.size, complexity, etc.) and are the independent variables of a prediction model for BPMNmeasurement. We proposed a set of theoretical and empirically validated base measuresfor the internal attributes of BPMN models.
Those measures could help during the course of process design, to take correctiveactions based on the assessment of process models’ internal characteristics. They canalso provide a rationale for adopting best-practice modeling rules in order to enhancequality in process models.
10.2.2 Minor Contributions
10.2.2.1 A Taxonomy for Process Modeling Languages
To address the research topic of quality in process modeling, we needed to choose, ac-cording to certain criteria, a language suitable for the specification and assessment of thequality attributes in process models. For this purpose, is proposed a pragmatic taxonomyin section 2.5.1 to enable a grounded justification of the choice made.
10.2.2.2 A Survey on Process Modeling Languages
In the chapter 2 we surveyed and rank a set of semi-formal and formal/executable pro-cess modeling languages, according to the aforementioned taxonomy.
10.2.2.3 Detected Flaws in the BPMN 2.0 Metamodel
In section 3.3.2, after BPMN metamodel analysis we detected a couple of flaws in thestandard specification and suggested some workarounds to face them.
10.2.2.4 Survey on BPMN Tools Maturity
Since, in this dissertation, we were concerned mainly with the quality aspects of processmodeling, we wanted to ascertain the effectiveness of current BPMN tools to ensure thequality of the generated models. We conclude in section 3.4.3 that none of the tools fromthe sample, fully detect the rules violations.
224
10. CONCLUSION 10.3. Future Work
10.2.2.5 BPMN Measurement Terminology
In section 6.2 we set up BPMN measurement terminology, grounded on concepts anddefinitions anchored upon the metrology vocabulary and the industry’s set of standardson quality, on measurement concepts from the discipline of Software Engineering, as wellas on an the SMO ontology.
10.2.2.6 A Framework for BPMN Measurement
In section 6.3 we proposed a framework for BPMN measurement which allows the cus-tomized setting up of guidelines for design and definition of sound BPMN measures.This framework was based upon the measure definition process GQM/MEDEA, a sys-tematic approach from software measurement.
10.2.2.7 BPMN Model-Based Testing Framework
The BPMN model-based testing framework presented in Figure 7.4 is a model-drivenengineering approach that guides the design and execution of a set of test cases againstan enhanced version of the BPMN metamodel with well-formedness rules.
10.2.2.8 BPMN Empirical Studies
In chapters 8 and 9 we conducted two sets of empirical studies using BPMN models.In those studies were tested several hypotheses regarding the quality of BPMN modelssamples.
Using the Binary Logistic Regression model, we were able to demonstrate its suitabil-ity for estimating the probability of a given BPMN model being faulty, giving as inputthe measures of its internal attributes.
10.3 Future Work
The future work intended to be done, for development of the paths traversed by thisdissertation, includes the following:
1. BPMN patterns and anti-patterns – Continue the evolution of the catalog of BPMNcommon modeling errors (see a sample in Appendix B), based on current and newversions of the BPMN standard. This catalog depicts about one hundred BPMNmodel snippets with rules violations together with the model counterpart withouterrors. Those BPMN model snippets were the test cases used in the framework forBPMN model-based testing, through which we tested the OCL invariants and opera-tions implemented in the BPMN metamodel.
2. BPMN standard – Contribute to the OMG’s future version of BPMN standard,with a document based on the above mentioned catalog, as well as with the for-malization of other rules that the community consider important. We expected by
225
10. CONCLUSION
this way, to contribute for the uniformization of the BPMN rules interpretation byBPMN tool makers, as well as for an improvement in the verification capabilities ofBPMN modeling tools.
3. Technology – Build an open source tool that allows, in practice, the well-formednessrules or best-practices rules reinforcement. This tool should provide hints (a de-piction of a model with the error found and its corrected version, as in AppendixB), guiding the process modelers in the overcoming of non-compliances in BPMNmodels.
4. New Empirical Studies – Conduct empirical studies for measuring the effects ofthe usage of the BPMN model checking tool on BPMN language’s learning curve.The studies should prioritize BPMN rules and constructs by clusters and give majorrelevance to the most used clusters of BPMN elements (see section 8.4.3). We expectthat a tool that implements the BPMN metamodel with the well-formedness rulesembedded, can contribute to decrease the modelers’ learning curve.
5. Generalization of Approach – Conduct studies regarding other external attributesof BPMN models (e.g. understandability) using the same approach as the one pre-sented in this dissertation for correctness.
6. Process-Oriented Quality – Use the results of this dissertation, regarding correct-ness of BPMN models, to support a new research line about quality upon BPMNprocess modeling, in a process-oriented perspective. The general ideas concerningthis research work we intend to develop in future are presented in Appendix H.
226
Bibliography
[Aab96] Anthony A. Aaby. Introduction to Programming Languages. http:
//www.emu.edu.tr/aelci/Courses/D-318/D-318-Files/
plbook/, Dec 1996.
[Aca13] AcademyHealth. Health Services Research Methods and Techniques.http://www.hsrmethods.org/glossary.aspx, Feb. 2013.
[AH10] Wil van der Aalst and Arthur ter Hofstede. Workflow Patterns homepage. http://www.workflowpatterns.com/, Dec. 2010.
[AL06] Krishna B. Athreya and Soumendra N. Lahiri. Measure Theory and Proba-bility Theory. Springer Texts in Statistics. Springer Science+Business Me-dia, LLC, 2006.
[All10] Thomas Allweyer. BPMN 2.0: Introduction to the Standard for BusinessProcess Modeling. Herstellung and Verlag: Books on Demand GmbH,Norderstedt, 2010.
[ALMN99] David E. Avison, Francis Lau, Michael D. Myers, and Peter Axel Nielsen.Action research. Communications of the ACM, 42(1):94–97, 1999.
[Ant96] Annie I. Anton. Goal-Based Requirements Analysis. In Proceedings of theSecond International Conference on Requirements Engineering, pages 136–144. IEEE, 1996.
[ARGP06] E.R. Aguilar, F. Ruiz, F. García, and M. Piattini. Evaluation measures forbusiness process models. In Proceedings of the 2006 ACM symposium onApplied computing, pages 1567–1568. ACM, 2006.
[Awa07] Ahmed Awad. BPMN-Q: A Language to Query Business Processes. InProceedings of the EMISA, volume 119, pages 115–128, 2007.
[AWW11] Ahmed Awad, Matthias Weidlich, and Mathias Weske. Visually speci-fying compliance rules and explaining their violations for business pro-cesses. Journal of Visual Languages & Computing, 22(1):30–55, 2011.
[B12] Egon Börger. Approaches to modeling business processes: a critical anal-ysis of BPMN, workflow patterns and YAWL. Software & Systems Model-ing, 11(3):305–318, 2012.
[BAGaB11] Ankica Barišic, Vasco Amaral, Miguel Goulão, and Bruno Barroca. Howto reach a usable DSL? Moving toward a Systematic Evaluation. InProceedings of the 5th International Workshop on Multi-Paradigm Modeling(MPM’2011), 2011.
[BAM10] Sérgio Bryton, Fernando Brito e Abreu, and Miguel Monteiro. Reduc-ing Subjectivity in Code Smells Detection: Experimenting with the LongMethod. In Proceedings of the 7th International Conference on the Qual-ity of Information and Communications Technology (QUATIC’2010), volumeThematic Track on Quality in ICT Reengineering and Refactoring, pages337–342. IEEE Computer Society Press, 2010.
[BB01] Erwan Breton and Jean Bézivin. Process-centered model engineering.In Proceedings of the Fifth IEEE International Enterprise Distributed ObjectComputing Conference, 2001. EDOC’01, pages 179–182. IEEE, 2001.
[BBKK04] J. Bae, H. Bae, S.H. Kang, and Y. Kim. Automatic control of workflowprocesses using ECA rules. IEEE Transactions on Knowledge and Data En-gineering, 16(8):1010–1023, 2004.
[BBM96] V.R. Basili, L.C. Briand, and Walcélio L Melo. A validation of object-oriented design metrics as quality indicators. IEEE Transactions on Soft-ware Engineering, 22(10):751–761, 1996.
[BCK03] Len Bass, Paul Clements, and Rick Kazman. Software Architecture in Prac-tice. SEI Series in Software Engineering. Addison-Wesley Pearson Edu-cation, Boston, second edition, 2003.
[BCR94] Victor R. Basili, Gianluigi Caldiera, and H Dieter Rombach. The goalquestion metric approach. Encyclopedia of software engineering, 2:528–532,1994.
[BD00] Christie Bolton and Jim Davies. Activity graphs and processes. In Pro-ceedings of the Integrated Formal Methods, pages 77–96. Springer, 2000.
[BD12] Paolo Bocciarelli and Andrea D’Ambrogio. Automated performanceanalysis of business processes. In Proceedings of the 2012 Symposium
228
BIBLIOGRAPHY
on Theory of Modeling and Simulation-DEVS Integrative M&S Symposium,page 10. Society for Computer Simulation International, 2012.
[BDG+08] Paul Baker, Zhen Ru Dai, Jens Grabowski, Ina Schieferdecker, and ClayWilliams. Model-Driven Testing: Using the UML Testing Profile. Springer-Verlag, first edition, 2008.
[BeA01] Fernando Brito e Abreu. Using OCL to formalize object oriented metricsdefinitions. In Tutorial in 5th International ECOOP Workshop on Quanti-tative Approaches in Object-Oriented Software Engineering (QAOOSE 2001),2001.
[BeAPFC10] Fernando Brito e Abreu, Raquel Porciúncula, Jorge Freitas, and José Car-los Costa. Definition and Validation of Complexity Metrics for ITSMProcess Models. 7th International Conference on the Quality of Informationand Communications Technology (QUATIC’2010), pages 79–88, Sept. 2010.
[BEEM95] Lionel Briand, Khaled El Emam, and Sandro Morasca. Theoretical andempirical validation of software product measures. International SoftwareEngineering Research Network, Technical Report ISERN-95-03, 1995.
[BHR84] Stephen D Brookes, Charles AR Hoare, and Andrew W Roscoe. A the-ory of communicating sequential processes. Journal of the ACM (JACM),31(3):560–599, 1984.
[BMB96] Lionel C. Briand, Sandro Morasca, and Victor R. Basili. Property-basedsoftware engineering measurement. IEEE Transactions on Software Engi-neering, 22(1):68–86, 1996.
[BMB02] Lionel C. Briand, Sandro Morasca, and Victor R. Basili. An operationalprocess for goal-driven definition of measures. IEEE Transactions on Soft-ware Engineering, 28(12):1106–1125, 2002.
[BMNT05] Jorge Biolchini, P Gomes Mian, A Candida Cruz Natali, and G HortaTravassos. Systematic review in software engineering. TechnicalReport 05, System Engineering and Computer Science DepartmentCOPPE/UFRJ, 2005.
[BO10] Dominik Birkmeier and Sven Overhage. Is BPMN Really First Choicein Joint Architecture Development? An Empirical Study on the Us-ability of BPMN and UML Activity Diagrams for Business Users. InGeorge T. Heineman, Jan Kofron, and Frantisek Plasil, editors, Researchinto Practice – Reality and Gaps, volume 6093 of Lecture Notes in Com-puter Science, pages 119–134. Springer Berlin Heidelberg, 2010. http:
[BPG+04] Paolo Bresciani, Anna Perini, Paolo Giorgini, Fausto Giunchiglia, andJohn Mylopoulos. Tropos: An agent-oriented software developmentmethodology. Autonomous Agents and Multi-Agent Systems, 8(3):203–236,2004.
[BPJ+10] Lars Braubach, Alexander Pokahr, Kai Jander, Winfried Lamersdorf, andBirgit Burmeister. Go4Flex: Goal-Oriented Process Modelling. In Mo-hammed Essaaidi, Michele Malgeri, and Costin Badica, editors, Intelli-gent Distributed Computing IV, volume 315, pages 77–87. Springer, 2010.
[BPM04] BPMI. Business Process Modeling Notation (BPMN) v1.0, May 2004.
[BPM11] OMG BPMN2. Business Process Model and Notation (BPMN) v2.0, Jan-uary 2011.
[BPV05] Jan van Bon, Mike Pieper, and Annelies van der Veen. Foundations of ITService Management: Based on ITIL, ITIL Version 2. Van Haren Publishing,2005.
[Bru04] Warren Brussee. Statistics for Six Sigma Made Easy! McGraw-Hill, 2004.
[BRvU00] Jörg Becker, Michael Rosemann, and Christoph von Uthmann. Guide-lines of business process modeling. In Business Process Management,pages 241–262. Springer, 2000.
[BS11] E. Börger and O. Sörensen. BPMN core modeling concepts: Inheritance-based execution semantics. In Handbook of Conceptual Modeling, pages287–332. Springer, 2011.
[BT08] Egon Börger and Bernhard Thalheim. A Method for Verifiable and Validat-able Business Process Modeling, volume 5316 of Lecture Notes in ComputerScience – Advances in Software Engineering. Springer Berlin / Heidelberg,2008.
[BV10] M.F. Bertoa and A. Vallecillo. Quality attributes for software metamod-els. Proceedings of the 13th TOOLS Workshop on Quantitative Ap-proaches in Object-Oriented Software Engineering (QAOOSE 2010), July2010. Málaga, Spain.
[BW10] Jeremy W. Bryans and Wei Wei. Formal analysis of bpmn models us-ing event-b. In Stefan Kowalewski and Marco Roveri, editors, FormalMethods for Industrial Critical Systems, volume 6371 of Lecture Notes inComputer Science, pages 33–49. Springer Berlin Heidelberg, 2010. http://dx.doi.org/10.1007/978-3-642-15898-8_3.
[Car07] Jorge Cardoso. Business process quality metrics: log-based complexityof workflow patterns. In On the Move to Meaningful Internet Systems 2007:CoopIS, DOA, ODBASE, GADA, and IS, pages 427–434. Springer, 2007.
[CBeA09] Anacleto Correia and Fernando Brito e Abreu. Integrating IT servicemanagement within the Enterprise Architecture. In Proceedings of theFourth International Conference on Software Engineering Advances, pages553–558. IEEE, 2009.
[CBeA10a] Anacleto Correia and Fernando Brito e Abreu. Model-Driven ServiceLevel Management. Phd research plan, FCT/UNL, September 2010.
[CBeA10b] Anacleto Correia and Fernando Brito e Abreu. Model-driven servicelevel management. In Mechanisms for Autonomous Management of Net-works and Services, pages 85–88. Springer, 2010.
[CBeA12] Anacleto Correia and Fernando Brito e Abreu. Adding preciseness toBPMN Models. In Proceedings of the 4th Conference on ENTERprise Infor-mation Systems (CENTERIS’2012), volume 5 of Procedia Technology, pages407–417. Elsevier, 2012.
[CBeAA11a] Anacleto Correia, Fernando Brito e Abreu, and Vasco Amaral. SLALOM:a language for Service Level Agreement specification and monitoring. InProceedings on the 3th INForum. Universidade de Coimbra, 2011.
[CBeAA11b] Anacleto Correia, Fernando Brito e Abreu, and Vasco Amaral. SLAME:A Service Level Agreements Method for Elicitation. In Proceedings on theCAPSI’2011, 2011.
[CC97] Ronald Christensen and R Christensen. Log-Linear Models and LogisticRegression. Springer Texts in Statistics. Springer-Verlag New York, Inc.,second edition, 1997.
[CdO13] Andre L. N. Campos and Toacy Cavalcante de Oliveira. Software Pro-cesses with BPMN: An Empirical Analysis. In Proceedings of the Product-Focused Software Process Improvement, pages 338–341. Springer, 2013.http://dx.doi.org/10.1007/978-3-642-39259-7_29.
[Che76] Peter Pin-Shan Chen. The entity-relationship model - toward a unifiedview of data. ACM Trans. Database Syst., 1(1):9–36, 1976.
[CKO92] Bill Curtis, Marc I. Kellner, and Jim Over. Process modeling. Com-mun. ACM, 35(9):75–90, September 1992. http://doi.acm.org/10.1145/130994.130998.
[CL09] Lawrence Chung and Julio Cesar Prado Leite. On Non-Functional Re-quirements in Software Engineering. In Conceptual Modeling: Foundationsand Applications, pages 363–379. Springer-Verlag, 2009.
[CMBC93] Giovanni Chiola, Marco Ajmone Marsan, Gianfranco Balbo, and GianniConte. Generalized stochastic Petri nets: A definition at the net level andits implications. IEEE Transactions on Software Engineering, 19(2):89–107,1993.
[CMNR06] J. Cardoso, J. Mendling, G. Neumann, and H.A. Reijers. A Discourseon Complexity of Process Models. In J. Eder and S. Dustdar, editors,Proceedings of the Business Process Management 2006 Workshops, volumeLNCS 4103, pages 115–126, 2006.
[CN95] Lawrence Chung and Brian A Nixon. Dealing with non-functional re-quirements: three experimental studies of a process-oriented approach.In Proceedings of the 17th International Conference on Software Engineering(ICSE 1995), pages 25–25. IEEE, 1995.
[CNW81] Nigel Cross, John Naughton, and David Walker. Design method andscientific method. Design studies, 2(4):195–201, 1981.
[CNYM00] Lawrence Chung, Brian A. Nixon, Eric Yu, and John Mylopoulos. Non-Functional Requirements in Software Engineering. Kluwer Academic Pub-lishers, Boston, 2000.
[Cod70] E. F. Codd. A relational model of data for large shared data banks. Com-mun. ACM, 13(6):377–387, 1970.
[Cro00] Stephen Crouch. Process Modelling for Requirements Capture. PhD thesis,University Of Southampton, 2000.
[CT12] Michele Chinosi and Alberto Trombetta. BPMN: An introduction to thestandard. Computer Standards & Interfaces, 34(1):124–134, 2012.
[Dav93] Thomas H. Davenport. Process innovation: reengineering work through in-formation technology. Harvard Business School Press, 1993.
[DDDGB08] Gero Decker, Remco Dijkman, Marlon Dumas, and Luciano García-Bañuelos. Transforming BPMN diagrams into YAWL nets. In BusinessProcess Management, pages 386–389. Springer, 2008.
[DDO07] Remco M. Dijkman, Marlon Dumas, and Chun Ouyang. Formal Seman-tics and Analysis of BPMN Process Models. Technical report, Queens-land University of Technology, 2007. http://eprints.qut.edu.au/7115/.
[DDO08] Remco M Dijkman, Marlon Dumas, and Chun Ouyang. Semantics andanalysis of business process models in BPMN. Information and SoftwareTechnology, 50(12):1281–1294, 2008.
[DFL91] Anne Dardenne, Stephen Fickas, and Axel van Lamsweerde. Goal-directed concept acquisition in requirements elicitation. In Proceedingsof the 6th international workshop on Software specification and design, pages14–21. IEEE Computer Society Press, 1991.
[dFS08] Pedro M. Gonzalez del Foyo and José Reinaldo Silva. Using time Petrinets for modelling and verification of timed constrained workflow sys-tems. In Proceedings of the ABCM Symposium Series in Mechatronics, vol-ume 3, pages 471–478, 2008.
[DGMR03] Islay Davies, Peter Green, Simon Milton, and Michael Rosemann. Usingmeta models for the comparison of ontologies. In Proceedings of the EighthCAiSE/IFIP8.1 International Workshop on Evaluation of Modeling Methods inSystems Analysis and Design, pages 16–17, 2003.
[DHP05] G. Dallons, P. Heymans, and I. Pollet. A Template-based Analysis ofGRL. In Proceedings of the EMMSAD, volume 5, pages 493–504, 2005.
[Dij82] E. W. Dijkstra. EWD 447: On the role of scientific thought. SelectedWritings on Computing: A Personal Perspective, pages 60–66, 1982.http://www.cs.utexas.edu/users/EWD/transcriptions/
EWD04xx/EWD447.html.
[DP11] K. Decreus and G. Poels. A goal-oriented requirements engineeringmethod for business processes. Information Systems Evolution, pages 29–43, 2011.
[DR98] Jörg Desel and Wolfgang Reisig. Place/transition petri nets. In WolfgangReisig and Grzegorz Rozenberg, editors, Lectures on Petri Nets I: BasicModels, volume 1491 of Lecture Notes in Computer Science, pages 122–173.Springer Berlin Heidelberg, 1998.
[DRC+06] N. Debnath, D. Riesco, M. P. Cota, J. B. Garcia Perez-Schofield, andD. R. M. Uva. Supporting the SPEM with a UML Extended WorkflowMetamodel. In Proceedings of the IEEE International Conference on Com-puter Systems and Applications (AICCSA), pages 1151–1154, 2006.
[DS02] Morris H. DeGroot and Mark J. Schervish. Probability and Statistics.Addison-Wesley, Pearson Education, Inc., fourth edition, 2002.
[DSP09] K. Decreus, M. Snoeck, and G. Poels. Practical Challenges for MethodsTransforming i* Goal Models into Business Process Models. In Proceed-ings of the 17th IEEE International Requirements Engineering Conference. RE’09., pages 15–23. IEEE, 2009.
[DVDA04] J. Dehnert and W.M.P. Van Der Aalst. Bridging the gap between businessmodels and workflow specifications. International Journal of CooperativeInformation Systems, 13(03):289–332, 2004.
[DVDATH05] M. Dumas, W. Van Der Aalst, and A. Ter Hofstede. Process-aware infor-mation systems. Wiley Online Library, 2005.
[DvLF93] Anne Dardenne, Axel van Lamsweerde, and Stephen Fickas. Goal-directed requirements acquisition. Science of Computer Programming,20(1):3–50, 1993.
[Ecl11] Eclipse. Atlas Transformation Language (ATL) v3.2.0. http://www.
eclipse.org/atl/, 2011.
[EFLR98] Andy Evans, Robert France, Kevin Lano, and Bernhard Rumpe. De-veloping the UML as a formal modelling notation. In Proceedings of theUML’98, LNCS, volume 1618, 1998.
[Erl07] Thomas Erl. SOA: Principles of Service Design. Service-Oriented Comput-ing Series. Prentice Hall, 2007.
[FABD10] Corradini Flavio, Polzonetti Alberto, Re Barbara, and Falcioni Damiano.An eclipse plug-in for formal verification of bpmn processes. In Proceed-ings of the Communication Theory, Reliability, and Quality of Service (CTRQ),2010 Third International Conference on, pages 144–149. IEEE, 2010.
[Fah08] Dirk Fahland. Translating UML2 Activity Diagrams to Petri nets for an-alyzing IBM WebSphere Business Modeler process models. Informatik-Berichte 226, Humboldt-Universität zu Berlin, 2008.
[Fav04] Jean-Marie Favre. Foundations of meta-pyramids: languages vs. meta-models – Episode II. Story of Thotus the Baboon. Language Engineeringfor Model-Driven Software Development, 4101, 2004.
[Fav05] Jean-Marie Favre. Foundations of Model (Driven) (Reverse) Engineer-ing : Models – Episode I: Stories of The Fidus Papyrus and of The So-larus. In Jean Bezivin and Reiko Heckel, editors, Proceedings of the Lan-guage Engineering for Model-Driven Software Development. InternationalesBegegnungs- und Forschungszentrum für Informatik (IBFI), SchlossDagstuhl, Germany, 2005.
[FCBeA08] Jorge Freitas, Anacleto Correia, and Fernando Brito e Abreu. An On-tology for IT Services. In Proceedings of the 13th Conference on SoftwareEngineering and Databases (JISBD’2008), 2008.
[FELR98] Robert France, Andy Evans, Kevin Lano, and Bernhard Rumpe. TheUML as a formal modeling notation. Computer Standards & Interfaces,19(7):325–334, 1998.
[FFK+11] Dirk Fahland, Cédric Favre, Jana Koehler, Niels Lohmann, HagenVölzer, and Karsten Wolf. Analysis on demand: Instantaneous sound-ness checking of industrial business process models. Data & KnowledgeEngineering, 70(5):448–466, 2011.
[Fie09] Andy Field. Discovering Statistics Using SPSS, Second Edition. SAGE Pub-lications Ltd, third edition, 2009.
[Fle10] Albert Fleischmann. What Is S-BPM?, volume 85 of Communications inComputer and Information Science, pages 85–106. Springer Berlin Heidel-berg, 2010.
[FMS11] Sanford Friedenthal, Alan Moore, and Rick Steiner. A Practical Guide toSysML: The Systems Modeling Language. Morgan Kaufmann, 2nd edition,2011.
[FN05] Jean-Marie Favre and Tam Nguyen. Towards a megamodel to modelsoftware evolution through transformations. Electronic Notes in Theoreti-cal Computer Science, 127(3):59–74, 2005.
[FP98] N.E. Fenton and S.L. Pfleeger. Software metrics: a rigorous and practicalapproach. PWS Publishing Co., 1998.
[FPMT01] A. Fuxman, M. Pistore, J. Mylopoulos, and P. Traverso. Model checkingearly requirements specifications in Tropos. In Proceedings of the FifthIEEE International Symposium on Requirements Engineering,, pages 174–181. IEEE, 2001.
[FPPR12] Damiano Falcioni, Andrea Polini, Alberto Polzonetti, and Barbara Re.Direct verification of BPMN processes through an optimized unfoldingtechnique. In Proceedings of the 12th International Conference on QualitySoftware (QSIC 2012), pages 179–188. IEEE, 2012.
[Fra03] David S. Frankel. BPM and MDA: The Rise of Model-DrivenEnterprise Systems. www.bptrends.com/publicationfiles/
06-03WPBPMandMDAWhitepaperFrankel11.pdf, June 2003.
[FSS+12a] Albert Fleischmann, Werner Schmidt, Christian Stary, Stefan Obermeier,and Egon Börger. From Language Acquisition to Subject-Oriented Model-ing Subject-Oriented Business Process Management, pages 9–23. SpringerBerlin Heidelberg, 2012.
[FSS+12b] Albert Fleischmann, Werner Schmidt, Christian Stary, Stefan Obermeier,and Egon Börger. S-BPM Method by Comparison Subject-Oriented BusinessProcess Management, pages 269–291. Springer Berlin Heidelberg, 2012.
[Ga08] Miguel Goulão. Component-Based Software Engineering: a Quantitative Ap-proach. PhD thesis, FCT/UNL, 2008.
[Gao06] Yi Gao. BPMN-BPEL Transformation and Round Trip Engineering.Technical report, eClarus Software, 2006. http://eclarus.com/
resources/BPMN_BPEL_Mapping.pdf.
[GBC+06] Félix García, Manuel F. Bertoa, Coral Calero, Antonio Vallecillo, Fran-cisco Ruíz, Mario Piattini, and Marcela Genero. Towards a consistentterminology for software measurement. Information and Software Technol-ogy, 48(8):631–644, 2006.
[GBR07] Martin Gogolla, Fabian Buttner, and Mark Richters. USE: A UML-basedspecification environment for validating UML and OCL. Science of Com-puter Programming, pages 69:27–34, 2007.
[GD13] Pieter Van Gorp and Remco Dijkman. A visual token-based formaliza-tion of BPMN 2.0 based on in-place transformations. Information andSoftware Technology, 55(2):365–394, 2013.
[GH13] B. Galloway and G. P. Hancke. Introduction to Industrial Control Net-works. Communications Surveys & Tutorials, IEEE, 15(2):860–880, 2013.
[Gia01] George M. Giaglis. A taxonomy of business process modeling and in-formation systems modeling techniques. International Journal of FlexibleManufacturing Systems, 13(2):209–228, 2001.
[GKMP04] Paolo Giorgini, Manuel Kolp, John Mylopoulos, and Marco Pistore. TheTropos Methodology Methodologies and Software Engineering for Agent Sys-tems, volume 11 of Multiagent Systems, Artificial Societies, and SimulatedOrganizations, pages 89–106. Springer US, 2004.
[GL06] Volker Gruhn and Ralf Laue. Complexity metrics for business processmodels. In Proceedings of the 9th international conference on business infor-mation systems (BIS 2006), volume Lecture Notes in Informatics, 85, 2006.
[GL07] Volker Gruhn and Ralf Laue. What business process modelers can learnfrom programmers. Science of Computer Programming, 65(1):4–13, 2007.
[GMS05] P. Giorgini, J. Mylopoulos, and R. Sebastiani. Goal-oriented require-ments analysis and reasoning in the Tropos methodology. EngineeringApplications of Artificial Intelligence, 18(2):159–171, 2005.
[GR05] P. Green and M. Rosemann. Ontological analysis of business sys-tems analysis techniques: experiences and proposals for an enhancedmethodology. Business systems analysis with ontologies, pages 1–27, 2005.
[GRI05] P.F. Green, M. Rosemann, and M. Indulska. Ontological evaluation ofenterprise systems interoperability using ebXML. IEEE Transactions onKnowledge and Data Engineering, 17(5):713–725, 2005.
[Gro07] Alexander Grosskopf. xBPMN - Formal Control Flow Specification of aBPMN based Process Execution Language. PhD thesis, Hasso-Plattner-Institute, 2007.
[GT09] D. Gagne and A. Trudel. Time-BPMN. In Proceedings of the IEEE Con-ference on Commerce and Enterprise Computing, CEC ’09, pages 361–367,2009.
[GW00] Ignacio E Grossmann and Arthur W Westerberg. Research challenges inprocess systems engineering. AIChE Journal, 46(9):1700–1703, 2000.
[Har04] P. Harmon. The OMG’s Model Driven Architecture and BPM,May 2004. http://www.bptrends.com/publicationfiles/
05-04NLMDAandBPM.pdf.
[Har07] P. Harmon. Business Process Change: a guide for business managers and BPMand Six Sigma Professionals. Morgan Kaufmann, 2007.
[HBS00] Adnan Hassan, Mohd Shariff Nabi Baksh, and Awaluddin M. Sha-haroun. Issues in quality engineering research. International Journal ofQuality & Reliability Management, 17(8):858–875, 2000.
[HC93] Michael Hammer and James Champy. Reengineering the Corporation: AManifesto for Business Revolution. Harperbusiness, Brealey, London, 1993.
[HCKP09] J.B. Hill, M. Cantara, M. Kerremans, and D.C. Plummer. Magic quadrantfor business process management suites. Gartner Research, 164485, 2009.
[HCV10] Peter Heisig, P. John Clarkson, and Sandor Vajna. Modelling and Man-agement of Engineering Processes. Springer-Verlag, London, 1st edition,2010.
[Hef04] Jeff Heflin. An Introduction to the OWL Web Ontology Language, 2004.
[HKP11a] R. Heinrich, A. Kappe, and B. Paech. Modeling Quality Informationwithin Business Process Models. In Proceedings of the 4th SQMB Work-shop, TUM-I1104, pages 4–13, 2011.
[HKP11b] R. Heinrich, A. Kappe, and B. Paech. Tool Support for the Comprehen-sive Modeling of Quality Information within Business Process Models.In Proceedings of the Enterprise Modelling and Information Systems Architec-tures (EMISA 2011), page 213, 2011.
[HMVHR06] J. Hernandez-Matias, A. Vizan, A. Hidalgo, and J. Rios. Evaluation oftechniques for manufacturing process analysis. Journal of Intelligent Man-ufacturing, 17(5):571–583, 2006.
[Hoa78] Charles Antony Richard Hoare. Communicating sequential processes.Communications of the ACM, 21(8):666–677, 1978.
[Hoa04] C. A. R. Hoare. Communicating Sequential Processes. Prentice Hall Inter-national, 2004.
[Hob81] Eric Hobsbawm. The Age of Revolution: Europe 1789–1848. Little, BrownBook Group Limited, 1981.
[Hol09] Jon Holt. A Pragmatic Guide to Business Process Modelling. BCS - BritishComputer Society, Swindon SN2 1FA, UK, 2nd edition, 2009.
[HOS06] Kees van Hee, Olivia Oanea, and Natalia Sidorova. Colored Petri Netsto Verify Extended Event-Driven Process Chains. In Proceedings of the 4thWorkshop on Modelling, Simulation, Verification and Validation of EnterpriseInformation Systems (MSVVEIS06), pages 76–85, 2006.
[HPSVH03] I. Horrocks, P.F. Patel-Schneider, and F. Van Harmelen. From SHIQ andRDF to OWL: The making of a web ontology language. Web semantics:science, services and agents on the World Wide Web, 1(1):7–26, 2003.
[HR85] Frederick Hayes-Roth. Rule-based systems. Commun. ACM, 28(9):921–932, 1985.
[HW02] James K. Huggins and Charles Wallace. An Abstract State MachinePrimer. Technical report, Computer Science Department, MichiganTechnological University, 4 December 2002.
[HW11] Paul Harmon and Celia Wolf. Business Process Modeling Survey, De-cember 2011. BPTrends.
[IEE98] IEEE. Std 1061-1998 – IEEE Standard for a Software Quality MetricsMethodology, 1998.
238
BIBLIOGRAPHY
[IMR09] Marta Indulska, Michael zur Muehlen, and Jan Recker. Measur-ing Method Complexity: The Case of the Business Process Mod-eling Notation. Technical report, BPM Center Report, Apr 2009.www.BPMcenter.org.
[ISO00] ISO. ISO 9004:2000 - Quality management systems – Guidelines for per-formance improvements, 2000. International Organization for Standard-ization.
[ISO03] ISO. ISO/IEC TR 14143-3:2003 – Information technology – Softwaremeasurement – Functional size measurement – Part 3: Verification offunctional size measurement methods, 2003. International Organizationfor Standardization.
[ISO04a] ISO. ISO/IEC 15504-4:2004 – Information technology – Process assess-ment – Part 4: Guidance on use for process improvement and processcapability determination, 2004. International Organization for Standard-ization.
[ISO04b] ISO. ISO/IEC 90003:2004 - Software engineering – Guidelines for theapplication of ISO 9001:2000 to computer software, 2004. InternationalOrganization for Standardization.
[ISO05a] ISO. ISO 9000-1:2005 - Quality management systems – Fundamentalsand vocabulary, 2005. International Organization for Standardization.
[ISO05b] ISO. ISO/IEC 25000:2005 – Software Engineering – Software productQuality Requirements and Evaluation (SQuaRE) – Guide to SQuaRE,2005. International Organization for Standardization.
[ISO07a] ISO. ISO/IEC 15939:2007 – Systems and software engineering – Mea-surement process, 2007. International Organization for Standardization.
[ISO07b] ISO. ISO/IEC Guide 99:2007, International vocabulary of metrology –Basic and general concepts and associated terms (VIM), 2007. Interna-tional Organization for Standardization.
[ISO08] ISO. ISO 9001:2008 - Quality management systems – Requirements, 2008.International Organization for Standardization.
[ISO11] ISO/IEC. Systems and software engineering – Systems and softwareQuality Requirements and Evaluation (SQuaRE) – System and softwarequality models, 2011-03-01 2011.
[Jar01] Richard D. Jarrard. Scientific methods. Technical report, Dept. of Ge-ology and Geophysics, University of Utah, 2001. www.iibhg.ukim.
[Jen94] Kurt Jensen. An introduction to the theoretical aspects of Coloured Petrinets. A decade of Concurrency Reflections and Perspectives, pages 230–272,1994.
[Jen97] Kurt Jensen. A brief introduction to Coloured Petri Nets. Tools and Algo-rithms for the Construction and Analysis of Systems, pages 203–208, 1997.
[Jen98] Kurt Jensen. An introduction to the practical use of Coloured Petri Nets.Lectures on Petri Nets II: Applications, pages 237–292, 1998.
[Joh08] Jendrik Johannes. ATL Use Case - Model Driven Performance Engineer-ing: From UML/SPT to AnyLogic. http://www.eclipse.org/m2m/atl/usecases/UML2AnyLogic/, 2008.
[Jon94] Christopher R. Jones. Improving Your Key Business Processes. The TQMMagazine, 6(2):25 – 29, 1994.
[JP05] Andreas Jedlitschka and Dietmar Pfahl. Reporting guidelines for con-trolled experiments in software engineering. In Proceedings of the 4thInternational Symposium on Empirical Software, pages 95–104. IEEE Com-puter Society, 17-18 Nov. 2005 2005.
[Jur11] Marko Jurišic. Transition between process models (BPMN) and servicemodels (WS-BPEL and other standards): A systematic review. Journal ofInformation and Organizational Sciences, 35(2):163–171, 2011.
[Kav02] E. Kavakli. Goal-oriented requirements engineering: A unifying frame-work. Requirements Engineering, 6(4):237–251, 2002.
[KCJ98] Lars M. Kristensen, Soren Christensen, and Kurt Jensen. The practi-tioner’s guide to coloured Petri nets. International Journal on SoftwareTools for Technology Transfer (STTT), 2(2):98–132, 1998.
[Kin03] Ekkart Kindler. On the semantics of EPCs: A framework for resolvingthe vicious circle. Technical report, Computer Science Department, Uni-versity of Paderborn, Germany, August 2003.
[Kit04] Barbara Kitchenham. Procedures for performing systematic reviews.Keele, UK, Keele University, 33:2004, 2004.
[KK97a] Peter Kueng and Peter Kawalek. Goal-based business process models:creation and evaluation. Business Process Management Journal, 3(1):17 –38, 1997.
[KK97b] Peter Kueng and Peter Kawalek. Goal-based business process models:creation and evaluation. Business Process Management Journal, 3(1):17 –38, 1997.
[KKGL10] S. Kühne, H. Kern, V. Gruhn, and R. Laue. Business process modelingwith continuous validation. Journal of Software Maintenance and Evolution:Research and Practice, 22(6–7):547–566, 2010.
[KL04] Evangelia Kavakli and Pericles Loucopoulos. Goal Driven RequirementsEngineering: Analysis and Critique of Current Methods, pages 102 – 124.IDEA Group, 2004.
[KL07] Birgit Korherr and Beate List. Extending the epc and the bpmn withbusiness process goals and performance measures. In ICEIS (3), pages287–294, 2007.
[KLS95] J. Krogstie, O.I. Lindland, and G. Sindre. Defining quality aspects forconceptual models. Proceedings of the IFIP8, 1:28–30, 1995.
[Ko09] R.K.L. Ko. A computer scientist’s introductory guide to business processmanagement (BPM). Crossroads, 15(4):4, 2009.
[Kor08] B. Korherr. Business Process Modelling: Languages, Goals, and Variabilities.VDM Publishing, 2008.
[KPF95] B. Kitchenham, S. L. Pfleeger, and N. Fenton. Towards a framework forsoftware measurement validation. Software Engineering, IEEE Transac-tions on, 21(12):929–944, 1995.
[KPP+02] Barbara A Kitchenham, Shari Lawrence Pfleeger, Lesley M Pickard, Pe-ter W Jones, David C. Hoaglin, Khaled El Emam, and Jarrett Rosenberg.Preliminary guidelines for empirical research in software engineering.IEEE Transactions on Software Engineering, 28(8):721–734, 2002.
[Kro03] J. Krogstie. Evaluating UML using a generic quality framework. Irm Press,2003.
[KS00] John Krogstie and Arne Solvberg. Information Systems Engineering: Con-ceptual Modeling in a quality perspective. Andersen Consulting / The Nor-wegian University of Science and Technology, 2000.
[KSJ06] J. Krogstie, G. Sindre, and H. Jørgensen. Process models representingknowledge for action: a revised quality framework. European Journal ofInformation Systems, 15(1):91–102, 2006.
[KSSB11] A. Kossiakoff, W.N. Sweet, S.J. Seymour, and S.M. Biemer. Systems Engi-neering Principles And Practice. Wiley Online Library, 2011.
[LA94] Frank Leymann and Wolfgang Altenhuber. Managing business pro-cesses as an information resource. IBM Systems Journal, 33(2):326–348,1994.
241
BIBLIOGRAPHY
[Lak95] Charles Lakos. From coloured Petri nets to object Petri nets. Applicationand Theory of Petri Nets 1995, pages 278–297, 1995.
[Lan03] Guy Lander. What is Sarbanes-Oxley? McGraw-Hill, 2003.
[lEE90] lEEE. Std 610.12-1990 – IEEE Standard Glossary of Software EngineeringTerminology, 1990.
[LK01] A.M. Latva-Koivisto. Finding a complexity measure for business pro-cess models. Helsinki University of Technology, Systems Analysis Laboratory,2001.
[LK06] Beate List and Birgit Korherr. An evaluation of conceptual business pro-cess modelling languages. In Proceedings of the 2006 ACM symposium onApplied computing, pages 1532–1539. ACM, 2006.
[LL05] Kenneth C. Laudon and Jane P. Laudon. Management Information Systems:Managing the Digital Firm. Prentice Hall, 9th edition, 2005.
[Lon04] Antoine Lonjon. Business process modeling and standardization. Tech-nical report, BPTrends, 2004. http://www.bptrends.com.
[LS97] Y. Lei and M.P. Singh. A comparison of workflow metamodels. In Pro-ceedings of the ER-97 Workshop on Behavioral Modeling and Design Transfor-mations: Issues and Opportunities in Conceptual Modeling, 1997.
[LS00] Sea Ling and Heinz Schmidt. Time Petri nets for workflow modellingand analysis. In Proceedings of the IEEE International Conference on Systems,Man, and Cybernetics, volume 4, pages 3039–3044. IEEE, 2000.
[LS07] Ruopeng Lu and Shazia Sadiq. A Survey of Comparative Business ProcessModeling Approaches Business Information Systems, volume 4439 of LectureNotes in Computer Science, pages 82–94. Springer Berlin / Heidelberg,2007.
[LSM12] Henrik Leopold, Sergey Smirnov, and Jan Mendling. On the refactor-ing of activity labels in business process models. Information Systems,37(5):443–459, 2012.
[LSS94] Odd Ivar Lindland, Guttorm Sindre, and Arne Solvberg. UnderstandingQuality in Conceptual Modeling. IEEE Softw., 11(2):42–49, 1994.
[LSW98] Peter Langner, Christoph Schneider, and Joachim Wehler. Petri Net BasedCertification of Event-Driven Process Chains Application and Theory of PetriNets 1998, volume 1420 of Lecture Notes in Computer Science edited by Desel,Jörg and Silva, Manuel, pages 286–305. Springer Berlin / Heidelberg, 1998.
[LvdA09] K.B. Lassen and W.M.P. van der Aalst. Complexity metrics for Workflownets. Information and Software Technology, 51(3):610–626, 2009.
[MADD04] D.A. Menasce, V.A.F. Almeida, L.W. Dowdy, and L. Dowdy. Performanceby design: computer capacity planning by example. Prentice Hall, 2004.
[MaP01] N. Melão and M. Pidd. A conceptual framework for understanding busi-ness processes and business process modelling. Information systems jour-nal, 10(2):105–129, 2001.
[Mar90] M. Marsan. Stochastic Petri Nets: an elementary introduction. Advancesin Petri Nets 1989, pages 1–29, 1990.
[May89] Richard E Mayer. Models for understanding. Review of educational re-search, 59(1):43–64, 1989.
[MB02] Stephen J. Mellor and Marc Balcer. Executable UML. A FoundationFor Model-Driven Architecture. Addison Wesley, 2002. http://www.
executableumlbook.com/.
[MB10] Marius Marusteri and Vladimir Bacarea. Comparing groups for statisti-cal differences: how to choose the right statistical test? Biochemia Medica,20(1):15–32, 2010.
[MCN92] J. Mylopoulos, L. Chung, and B. Nixon. Representing and using non-functional requirements: A process-oriented approach. Software Engi-neering, IEEE Transactions on, 18(6):483–497, 1992.
[MCY99] J. Mylopoulos, L. Chung, and E. Yu. From object-oriented to goal-oriented requirements analysis. Communications of the ACM, 42(1):31–37,1999.
[Men07] J. Mendling. Detection and Prediction of Errors in EPC Business ProcessModels. PhD thesis, Vienna University of Economics and Business Ad-ministration, 2007.
[MGSA10] G.M. Muketha, A.A.A. Ghani, M.H. Selamat, and R. Atan. A Surveyof Business Process Complexity Metrics. Information Technology Journal,9:1336–1344, 2010.
[MHO06] R. Matulevicius, P. Heymans, and A.L. Opdahl. Ontological analysisof KAOS using separation of reference. In Proceedings of the EMMSAD,volume 6, 2006.
[MHO07] R. Matulevicius, P. Heymans, and A. Opdahl. Comparing GRL andKAOS using the UEML Approach Enterprise Interoperability II, pages 77–88. Springer London, 2007.
[Mil56] G. A. Miller. The magical number seven, plus or minus two: Some lim-its on our capacity for processing information. Psychological Review, 63(2):81–97, 1956.
[Mil99] Robin Milner. Communicating and Mobile Systems: the π-Calculus. Cam-bridge University Press, 1999.
[MKC01] John Mylopoulos, Manuel Kolp, and Jaelson Castro. UML for Agent-Oriented Software Development: The Tropos Proposal, volume 2185 of LectureNotes in Computer Science, pages 422–441. Springer Berlin / Heidelberg,2001.
[MM97a] V. B. Misic and S. Moser. From formal metamodels to metrics: an object-oriented approach. In Proceedings of the Technology of Object-Oriented Lan-guages (TOOLS 1997), pages 330–339. IEEE, 1997.
[MM97b] S. Moser and V. B. Misic. Measuring class coupling and cohesion: aformal metamodel approach. In Proceedings of the Software EngineeringConference, 1997. Asia Pacific and International Computer Science Conference1997. APSEC ’97 and ICSC ’97, pages 31–40, 1997.
[MN05] Jan Mendling and Markus Nüttgens. EPC Markup Language (EPML)– An XML-Based Interchange Format for Event-Driven Process Chains(EPC). Technical report, Vienna University of Economics and BusinessAdministration, March 2005.
[MNvdA07] Jan Mendling, Gustaf Neumann, and Wil van der Aalst. Understandingthe Occurrence of Errors in Process Models Based on Metrics. In RobertMeersman and Zahir Tari, editors, On the Move to Meaningful Internet Sys-tems 2007: CoopIS, DOA, ODBASE, GADA, and IS, volume 4803 of LectureNotes in Computer Science, pages 113–130. Springer Berlin / Heidelberg,2007. 10.1007/978-3-540-76848-7 9.
[Moo05] D.L. Moody. Theoretical and practical issues in evaluating the quality ofconceptual models: current state and future directions. Data & KnowledgeEngineering, 55(3):243–276, 2005.
[Mor71] Charles William Morris. Writings on the General Theory of Signs. Moutonde Gruyter, The Hague, The Netherlands, 1971.
[MPSP+09] Boris Motik, Peter F Patel-Schneider, Bijan Parsia, Conrad Bock, AchilleFokoue, Peter Haase, Rinke Hoekstra, Ian Horrocks, Alan Ruttenberg,Uli Sattler, et al. OWL 2 web ontology language: Structural specificationand functional-style syntax. W3C recommendation, 27:17, 2009. http:
[MR03] Douglas C. Montgomery and George C. Runger. Applied Statistics andProbability for Engineers. John Wiley, third edition, 2003.
[MR08] M. Muehlen and J. Recker. How much language is enough? Theoreti-cal and practical use of the business process modeling notation. In Pro-ceedings of the Advanced information systems engineering, pages 465–479.Springer, 2008.
[MRC07] Jan Mendling, Hajo Reijers, and Jorge Cardoso. What Makes Process Mod-els Understandable? Business Process Management, volume 4714 of LectureNotes in Computer Science, pages 48–63. Springer Berlin / Heidelberg,2007.
[MRvdA10] J. Mendling, H.A. Reijers, and W.M.P. van der Aalst. Seven process mod-eling guidelines (7PMG). Information and Software Technology, 52(2):127–136, 2010.
[MSBS03] D. Moody, G. Sindre, T. Brasethvik, and A. Sølvberg. Evaluating thequality of process models: Empirical testing of a quality framework.Conceptual Modeling—ER 2002, pages 380–396, 2003.
[MSW10] A. Meneely, B. Smith, and L. Williams. Software metrics validation crite-ria: a systematic literature review. North Carolina State University Depart-ment of Computer Science, Raleigh, NC, pages 27695–8206, 2010.
[MSW12] Andrew Meneely, Ben Smith, and Laurie Williams. Validating softwaremetrics: A spectrum of philosophies. ACM Transactions on Software Engi-neering and Methodology (TOSEM), 21(4):24, 2012.
[MTJ+10] Hafedh Mili, Guy Tremblay, Guitta Bou Jaoude, Eric Lefebvre, LamiaElabed, and Ghizlane El Boussaidi. Business process modeling lan-guages: Sorting through the alphabet soup. ACM Computing Surveys(CSUR), 43(1):4, 2010.
[Mur89] T. Murata. Petri nets: Properties, analysis and applications. Proceedingsof the IEEE, 77(4):541–580, 1989.
[Myl98] J. Mylopoulos. Information Modeling in the Time of the Revolution.Information Systems, 23(3):127–155, 1998.
[Nat11] Christine Natschleger. Towards a BPMN 2.0 Ontology. Lecture Notes inBusiness Information Processing, 95:1–15, 2011.
[NBZ06] Nachiappan Nagappan, Thomas Ball, and Andreas Zeller. Mining met-rics to predict component failures. In Proceedings of the 28th internationalconference on Software engineering, ICSE ’06, pages 452–461, New York,
245
BIBLIOGRAPHY
NY, USA, 2006. ACM. http://doi.acm.org/10.1145/1134285.
1134349.
[OAS07] OASIS. Web Services Business Process Execution Language Version 2.0,11 April 2007 2007.
[OHNAEE+07] Javier Ortiz-Hernández, Erika M Nieto-Ariza, Hugo Estrada-Esquivel,Guillermo Rodríguez-Ortiz, and Azucena Montes-Rendon. A theoreticalevaluation for assessing the relevance of modeling techniques in busi-ness process modeling. In Proceedings of the Fourth International Workshopon Software Quality Assurance: in conjunction with the 6th ESEC/FSE jointmeeting, pages 102–107. ACM, 2007.
[OM98] Min Oh and II Moon. Framework of dynamic simulation for complexchemical processes. Korean Journal of Chemical Engineering, 15(3):231–242,1998.
[OMG03a] OMG. Common Warehouse Metamodel (CWM). Object ManagementGroup, 2003. http://www.omg.org/spec/CWM/1.1.
[OMG03b] OMG. MDA Guide Version 1.0.1. Object Management Group, 2003.http://www.omg.org/cgi-bin/doc?omg/03-06-01.
[OMG05] OMG. UML Profile for Schedulability, Performance, and Time (v1.1).Object Management Group, January 2005.
[OMG06] OMG. Object Constraint Language (OCL). Object Management Group,2006.
[OMG11] OMG. MOF 2.0/XMI Mapping, Version 2.4.1. Object ManagementGroup, August 2011. http://www.omg.org/spec/XMI/2.4.1/
PDF.
[OMG12a] OMG. Semantics of Business Vocabulary and Business Rules (SBVR).Object Management Group, 2012. http://www.omg.org/spec/
SBVR/1.1/index.htm.
[OMG12b] OMG. System Modeling Language (SysML). Object ManagementGroup, June 2012. http://www.omg.org/spec/SysML/1.3/.
[OMG13] OMG. Semantics of a Foundational Subset for Executable UML Models(FUML). Object Management Group, 2013. http://www.omg.org/
spec/FUML/1.1.
[Par72] David Lorge Parnas. On the criteria to be used in decomposing systemsinto modules. Communications of the ACM, 15(12):1053–1058, 1972.
[PCCW93] Mark C. Paulk, Bill Curtis, Mary Beth Chrissis, and Charles V. Weber.Capability Maturity Model for Software, Version 1.1. Technical report,SEI – Carnegie Mellon University, 1993.
[Per06] James R. Persse. Process Improvement Essentials. O’Reilly Media, Inc.,2006.
[Pet62] Carl A. Petri. Kommunikation mit Automaten. PhD thesis, Institut furInstrumentelle Mathematik, 1962.
[Pet77] James L. Peterson. Petri nets. ACM Computing Surveys (CSUR), 9(3):223–252, 1977.
[Pet81] J. L. Peterson. Petri Net Theory and the Modeling of Systems. Prentice HallPTR, 1981.
[PFS10] Elke Pulvermueller, Sven Feja, and Andreas Speck. Developer-friendly verification of process-based systems. Knowledge-Based Systems,23(7):667–676, 2010.
[Pha98] Keith Thomas Phalp. The CAP framework for business process mod-elling. Information and Software Technology, 40(13):731–744, 1998.
[PI92] R.S. Pressman and D. Ince. Software engineering: a practitioner’s approach,volume 5. McGraw-hill New York, 1992.
[Pop35] Karl Popper. The Logic of Scientific Discovery. Taylor & Francis e-Library,1935.
[PRP08] Jose Manuel Perez, Francisco Ruiz, and Mario Piattini. MDE for BPM:a systematic review. In Software and Data Technologies, pages 127–135.Springer, 2008.
[PS08] Viara Popova and Alexei Sharpanskykh. Process-oriented organisationmodelling and analysis. Enterprise Information Systems, 2(2):157–176,2008.
[PS09] Viara Popova and Alexei Sharpanskykh. Constraint-based modellingand analysis of organisations. In Proceedings of the 2009 ACM symposiumon Applied Computing, pages 283–284. ACM, 2009.
[PS10a] Viara Popova and Alexei Sharpanskykh. Modeling organizational per-formance indicators. Information Systems, 35(4):505–527, 2010.
[PS11] Viara Popova and Alexei Sharpanskykh. Formal modelling of organ-isational goals based on performance indicators. Data Knowl. Eng.,70(4):335–364, 2011.
[PS13] Susanne Patig and Manuela Stolz. A pattern-based approach for theverification of business process descriptions. Information and SoftwareTechnology, 55(1):58–87, 2013.
[PZ08] C.J. Pavlovski and J. Zou. Non-functional requirements in business pro-cess modeling. In Proceedings of the fifth Asia-Pacific conference on Concep-tual Modelling-Volume 79, pages 103–112. Australian Computer Society,Inc., 2008.
[RCG+09] E. Rolón, J. Cardoso, F. García, F. Ruiz, and M. Piattini. Analysis andValidation of Control-Flow Complexity Measures with BPMN ProcessModels. Enterprise, Business-Process and Information Systems Modeling,29:58–70, 2009.
[Rec07] J. C. Recker. Understanding Quality in Process Modelling: towards aholistic perspective. Australasian Journal of Information Systems, 14(2):43–63, 2007.
[REH11] Mohamed Ramadan, Hicham G Elmongui, and Riham Hassan. BPMNFormalisation using Coloured Petri Nets. In Proceedings of the 2nd GSTFAnnual International Conference on Software Engineering & Applications(SEA 2011), 2011.
248
BIBLIOGRAPHY
[RG02] Michael Rosemann and Peter Green. Developing a meta model forthe Bunge-Wand-Weber ontological constructs. Information Systems,27(2):75–91, 2002.
[RGI04] M. Rosemann, P. Green, and M. Indulska. A reference methodology forconducting ontological analyses. Conceptual Modeling-ER 2004, pages110–121, 2004.
[RI07] J.C. Recker and M. Indulska. An ontology-based evaluation of processmodeling with petri nets. IBIS-International Journal of Interoperability inBusiness Information Systems, 2(1):45–64, 2007.
[RIG07] Jan Recker, Marta Indulska, and Peter Green. Extending RepresentationalAnalysis: BPMN User and Developer Perspectives - Business Process Manage-ment, volume 4714 of Lecture Notes in Computer Science, pages 384–399.Springer Berlin / Heidelberg, 2007.
[RIRG05] J. Recker, M. Indulska, M. Rosemann, and P. Green. Do process mod-elling techniques get better? A comparative ontological analysis ofBPMN. In B. Campbell, J. Underwood, and D. Bunker, editors, Pro-ceedings of the 16th Australasian Conference on Information Systems, Sidney,Australia, 2005.
[RIRG06] J.C. Recker, M. Indulska, M. Rosemann, and P. Green. How good isBPMN really? Insights from theory and practice. In Proceedings of theEuropean Conference on Information Systems, volume 14, pages 1–12. ITUniversity of Gotteborg, 2006.
[RM98] M. Roseman and M. Muehlen. Evaluation of workflow managementsystems-a meta model approach. Australian Journal of Information Sys-tems, 6:103–116, 1998.
[RM06] J.C. Recker and J. Mendling. On the translation between BPMN andBPEL: Conceptual mismatch between process modeling language. InProceedings of the 18th International Conference on Advanced InformationSystems Engineering, Workshops and Doctoral Consortium, pages 521–532.Namur University Press, 2006.
[RPD98] John O. Rawlings, Sastry G. Pantula, and David A. Dickey. Applied Re-gression Analysis A Research Tool. Springer Texts in Statistics. Springer-Verlag New York, Inc., second edition, 1998.
[RRG+09] L. Reynoso, E. Rolón, M. Genero, F. García, F. Ruiz, and M. Piattini. For-mal definition of measures for BPMN models. Software Process and Prod-uct Measurement, 5891:285–306, 2009.
249
BIBLIOGRAPHY
[RRGI09] M. Rosemann, J. Recker, P.F. Green, and M. Indulska. Using ontology forthe representational analysis of process modelling techniques. Interna-tional Journal of Business Process Integration and Management, 4(4):251–265,2009.
[RRIG09] J.C. Recker, M. Rosemann, M. Indulska, and P. Green. Business processmodeling: a comparative analysis. Journal of the Association for Informa-tion Systems, 10(4):333–363, 2009.
[RRK07] J. Recker, M. Rosemann, and J. Krogstie. Ontology-versus pattern-basedevaluation of process modeling languages: a comparison. Communica-tions of the Association for Information Systems, 20(48):774–799, 2007.
[RW05] G. Regev and A. Wegmann. Where do goals come from: the underlyingprinciples of goal-oriented requirements engineering. In Proceedings ofthe 13th IEEE International Conference on Requirements Engineering, pages353–362. IEEE, 2005.
[Sal04] Mathias Sallé. IT Service Management and IT Governance: Review,Comparative Analysis and their Impact on Utility Computing. Technicalreport, Hewlett-Packard Research Labs, 2004.
[Sch06] Douglas C Schmidt. Model-driven engineering. Computer-IEEE Com-puter Society, 39(2):25, 2006.
[SD97] Graeme Shanks and Peta Darke. Quality in Conceptual Modelling: Link-ing Theory and Practice. In Proceedings of the PACIS 1997, volume Paper76, 1997.
[SEI10a] Software Engineering Institute SEI. CMMI for Acquisition, Version 1.3 -CMMI-ACQ, V1.3. Technical report, SEI – Carnegie Mellon University,2010.
[SEI10b] Software Engineering Institute SEI. CMMI for Development,Version 1.3- CMMI-DEV, V1.3. Technical report, SEI – Carnegie Mellon University,2010.
[SEI10c] Software Engineering Institute SEI. CMMI for Services, Version 1.3 -CMMI-SVC, V1.3. Technical report, SEI – Carnegie Mellon University,2010.
[SG05] Andrew Stellman and Jennifer Greene. Applied Software Project Manage-ment. O’Reilly Media, 2005.
[Sil09] Bruce Silver. BPMN Method and Style. Cody-Cassidy Press, Aptos, 1stedition, 2009.
[Sim96] Herbert Simon. Sciences of the Artificial. The MIT Press, Cambridge, 3rdedn edition, 1996.
[Sin06] G. Sindre. An analytical evaluation of BPMN using a semiotic qualityframework. Advanced topics in database research, 5:94, 2006.
[SLT09] M. Saunders, P. Lewis, and A. Thornhill. Research methods for businessstudents. Financial Times Prentice Hall, 5th edition, 2009.
[Smi08] David A. Smith. Implementing Metrics for IT Service Management. VanHaren Publishing, Zaltbommel, NL, 2008.
[SMJ00] Rick Sturm, Wayne Morris, and Mary Jander. Foundations of Service LevelManagement. Sams Publishing, 1st edition (april 15, 2000) edition, 2000.
[SN00] August-Wilhelm Scheer and Markus Nüttgens. ARIS Architecture andReference Models for Business Process Management. In Proceedings ofthe Business Process Management, Models, Techniques, and Empirical Studies,pages 376–389, London, UK, UK, 2000. Springer-Verlag. http://dl.
acm.org/citation.cfm?id=647778.734910.
[Spe81] Paul Spector. Research Designs. SAGE Publications, 1981.
[SRD03] Wasana Sedera, Michael Rosemann, and Gabriella Doebeli. A ProcessModelling Success Model: Insights from a Case Study. In Proceedings ofthe 11th European Conference on Information Systems, pages 1–11, Naples,Italy, 2003. ECIS. http://eprints.qut.edu.au/11129/.
[SSB01] Robert F Stärk, Joachim Schmid, and Egon Börger. Java and the Java Vir-tual Machine. Springer Heidelberg, 2001.
[STW03] G. Shanks, E. Tansley, and R. Weber. Using ontology to validate concep-tual models. Communications of the ACM, 46(10):85–89, 2003.
[Sup10] Basel Committee on Banking Supervision. Basel III: A global regulatoryframework for more resilient banks and banking systems, 2010.
[SWB+12] Robert Shapiro, Stephen A. White, Conrad Bock, Nathaniel Palmer,Michael zur Muehlen, Marco Brambilla, and Denis Gagné. BPMN 2.0Handbook. Future Strategies Inc., second edition edition, 2012.
[Tay11] Frederick Winslow Taylor. The principles of scientific management. Harper& Brothers, New York, London, 1911.
[tHvdAAR09] A.H.M. ter Hofstede, W.M.P. van der Aalst, M. Adams, and N. Russell.Modern Business Process Automation: YAWL and its support environment.Springer, 2009.
[Tro06] William M.K. Trochim. Research Methods Knowledge Base. http://
www.socialresearchmethods.net/kb/, 2006.
[Tsc10] Willi Tscheschner. Transformation from EPC to BPMN. Technical report,Oryx Research, 2010.
[UL06] Mark Utting and Bruno Legeard. Practical Model-Based Testing: A ToolsApproach. Morgan Kaufmann, 1rst edition, 2006.
[UoT12] Canada University of Toronto. GRL - Goal-oriented Requirement Lan-guage. http://www.cs.toronto.edu/km/GRL/, 2012.
[VCM+07] Irene Vanderfeesten, Jorge Cardoso, Jan Mendling, Hajo A. Reijers, andWil van der Aalst. Quality Metrics for Business Process Models, pages 179–190. Future Strategies Inc., FL, USA, 2007.
[VdA94] W. M. P. Van der Aalst. Putting high-level Petri nets to work in industry.Computers in Industry, 25(1):45–54, 1994.
[vdA96] Wil M. P. van der Aalst. Structural characterizations of sound workflow nets.Eindhoven University of Technology, Department of Mathematics andComputing Science, 1996.
[VdA97] Wil M. P. Van der Aalst. Verification of workflow nets, pages 407–426.Springer, 1997.
[vdA98] W.M.P. van der Aalst. The application of Petri nets to workflow manage-ment. Journal of circuits, systems, and computers, 8(01):21–66, 1998.
[vdA99] W. M. P. van der Aalst. Formalization and verification of event-drivenprocess chains. Information and Software Technology, 41(10):639–650, 1999.
[vDA00] W. van Der Aalst. Workflow verification: Finding control-flow errorsusing petri-net-based techniques. Business Process Management, pages161–183, 2000.
[vdABL08] Wil M. P. van der Aalst and Kristian Bisgaard Lassen. Translating un-structured workflow processes to readable BPEL: Theory and implemen-tation. Information and Software Technology, 50(3):131–159, 2008.
[VdAtHW03] W. Van der Aalst, A. ter Hofstede, and M. Weske. Business process man-agement: A survey. Business Process Management, pages 1019–1019, 2003.
[VdAVH96] W. M. P. Van der Aalst and K. M. Van Hee. Business process redesign: aPetri-net-based approach. Computers in Industry, 29(1):15–26, 1996.
[vDvdAV05] Boudewijn F van Dongen, Wil MP van der Aalst, and Henricus MW Ver-beek. Verification of EPCs: Using reduction rules and Petri nets. In Pro-ceedings of the Advanced Information Systems Engineering, pages 372–386.Springer, 2005.
[VL01] Axel Van Lamsweerde. Goal-Oriented Requirements Engineering: AGuided Tour. In Proceedings of the Fifth IEEE International Symposium onRequirements Engineering (RE ’01), pages 249–262. IEEE, 2001.
[vL09] Axel van Lamsweerde. Requirements Engineering. John Wiley & Sons Ltd,West Sussex, England, 2009.
[VRM+08] Irene Vanderfeesten, Hajo Reijers, Jan Mendling, Wil van der Aalst, andJorge Cardoso. On a Quest for Good Process Models: The Cross-ConnectivityMetric Advanced Information Systems Engineering, volume 5074 of LectureNotes in Computer Science, pages 480–494. Springer Berlin / Heidelberg,2008.
[vvdAS11] Wil van van der Aalst and Christian Stahl. Modeling Business Processes:A Petri Net-Oriented Approach. The MIT Press, 2011.
[Was04] Larry Wasserman. All of Statistics: A Concise Course in Statistical Inference.Springer Texts in Statistics. Springer Science+Business Media, Inc., 2004.
[Was06] Larry Wasserman. All of Nonparametric Statistics. Springer Texts in Statis-tics. Springer Science+Business Media, Inc., 2006.
[WDGW08] Matthias Weidlich, Gero Decker, Alexander Großkopf, and MathiasWeske. BPEL to BPMN: The Myth of a Straight-Forward Mapping, 2008.
[Wed06] Ian Wedgwood. Lean Sigma: A Practitioner’s Guide. Prentice Hall, 2006.
[Wes07] Mathias Weske. Business Process Management - Concepts, Languages, Ar-chitectures. Springer-Verlag Berlin Heidelberg, 2007.
[WHJC06] Wang Wei, Ding Hongwei, Dong Jin, and Ren Changrui. A Compari-son of Business Process Modeling Methods. In Proceedings of the IEEEInternational Conference on Service Operations and Logistics, and Informatics,SOLI ’06, pages 1136–1141, 2006.
[WM08] Stephen A. White and Derek Miers. BPMN Modeling and Reference Guide:Understanding and Using BPMN. Future Strategies, Inc., LighthousePoint, Florida, USA, 2008.
[WP93] W.D. Waltman and A. Presley. Reading & Critiquing an IDEF0 Model.Automation & Robotics Research. Institute, Texas, July 1993.
[WRH+00] C. Wohlin, P. Runeson, M. Höst, M.C. Ohlsson, B. Regnell, and A. Wess-lén. Experimentation in Software Engineering: An Introduction. KluwerAcademic Publishers, 2000.
[WW90a] Y. Wand and R. Weber. An ontological model of an information system.IEEE Transactions on Software Engineering, 16(11):1282–1292, 1990.
[WW90b] Y. Wand and R. Weber. Toward a theory of the deep structure of informa-tion systems. University of British Columbia, Faculty of Commerce andBusiness Administration, 1990.
[WW96] Y. Wand and R.Y. Wang. Anchoring data quality dimensions in ontolog-ical foundations. Communications of the ACM, 39(11):86–95, 1996.
[YM94a] E.S.K. Yu and J. Mylopoulos. Understanding ”why” in software processmodelling, analysis, and design. In Proceedings of the 16th internationalconference on Software engineering, pages 159–168. IEEE Computer SocietyPress, 1994.
[YM94b] E.S.K. Yu and J. Mylopoulos. Using goals, rules, and methods to sup-port reasoning in business process reengineering. In Proceedings of theTwenty-Seventh Hawaii International Conference on System Sciences, vol-ume 4, pages 234–243. IEEE, 1994.
[YM98] E. Yu and J. Mylopoulos. Why goal-oriented requirements engineering.In Proceedings of the 4th International Workshop on Requirements Engineer-ing: Foundations of Software Quality, pages 15–22, 1998.
[YML96] E.S.K. Yu, J. Mylopoulos, and Y. Lespérance. AI models for businessprocess reengineering. IEEE expert, 11(4):16–23, 1996.
[Yu97] E.S.K. Yu. Towards modelling and reasoning support for early-phaserequirements engineering. In Proceedings of the Third IEEE InternationalSymposium on Requirements Engineering, pages 226–235. IEEE, 1997.
[ZJ08] Li Zhang and Wei Jiang. Transforming business requirements into BPEL:A MDAa-based approach to web application development. In Proceed-ings of the IEEE International Workshop on Semantic Computing and Systems,WSCS’08, pages 61–66. IEEE, 2008.
[zM99] M. zur Muhlen. Evaluation of workflow management systems usingmeta models. In Proceedings of the 32nd Annual Hawaii International Con-ference on System Sciences. HICSS-32, page 11 pp. IEEE, 1999.
[zMR08] Michael zur Muehlen and Jan Recker. How much BPMNdo you need. http://www.bpm-research.com/2008/03/03/
how-much-bpmn-do-you-need, 2008.
[ZMRI07] M. Zur Muehlen, J. Recker, and M. Indulska. Sometimes less is more:Are process modeling languages overly complex? In Proceedings of theEleventh International IEEE EDOC Conference Workshop, EDOC’07, pages197–204. IEEE, 2007.
accessible population is the population that will be accessible to the researcher on theimpossibility of accessing the overall population [Tro06]. 158
assignment by cutoff is the assignment to groups done by a pragmatic method. It isuseful when those who allocate subjects may be subject to bias or favoritism [Tro06].159
boundedness A Petri net is bounded if, for a given initial marking, the number of tokensin any given place never exceeds a finite number k [MTJ+10]. 33
compensatory equalization occurs when different groups receive different treatments.So, the control group may receive some compensation due to the fact of not usingthe treatment being tested. One must be aware that if the compensation has aneffect on the performance of the control group, this may jeopardize the the overallexperiment. 196
compensatory rivalry occurs when subjects from the group not receiving a new treat-ment feel penalized. As reaction they could work harder than they normally wouldto counteract that effect. 196
confidence interval tells us, in statistical analysis, the reliability of the sample parame-ter as compared to the whole population parameter. It conveys an expect certaindegree of error and uncertainty, which depends on a variety of conditions, like thenumber of subjects in the experiment and the way they represent the whole popu-lation [Shu09]. 173
convenience sample is a type of non-probability sampling which involves the samplebeing drawn from that part of the population which is close to hand. That is, asample population selected because it is readily available and convenient. The re-searcher using such a sample cannot scientifically make generalizations about thetotal population from this sample because it would not be representative enough[Tro06]. 63, 158
259
Glossary
data independence It refers to the insusceptibility of user applications to make changesin the structure of data. Furthermore, physical data independence concerns withhiding the details of the storage structure from user applications. [Cod70]. 12
data set corresponds to the contents of a single database table, or a single statistical datamatrix, where each column of the table represents a particular variable, and eachrow corresponds to a given member of the data set in question. The data set listsvalues for each of the variables, for each member of the data set. Each value isknown as a datum. The data set may comprise data for one or more members,corresponding to the number of rows. 165
deadlock A deadlock is a set of places such that every transition which outputs to oneof the places in the deadlock also inputs from one of these places. This means thatonce all of the places in the deadlock become unmarked, the entire set of places willalways be unmarked; no transition can place a token in the deadlock because thereis no token in the deadlock to enable a transition which outputs to a place in thedeadlock [Pet81]. 33, 71, 75, 79
deductive reasoning also known as deduction, it is a kind of reasoning in which specificexamples are derived from general propositions. Deductive reasoning, starts witha general principle and deduces that it applies to a specific case . 149
dependent variable also called responding variable. This is the factor that is the outcomemeasure, i.e. the effect (or change). 101
descriptive research seeks to depict what already exists in a group or population. De-scriptive studies do not seek to measure the effect of a variable; they seek only todescribe [Tro06]. 159
empirical means information gained by experience, observation, or experiment. Thecentral theme in scientific method is that all evidence must be empirical whichmeans it is based on evidence. In scientific method the word empirical refers tothe use of working hypothesis that can be tested using observation and experiment[Shu09]. 75, 151, 203
falsifiability or refutability is the belief that for any hypothesis to have credence, it mustbe inherently disprovable before it can become accepted as a scientific hypothesisor theory [Shu09]. 149, 172
independent variable also called experimental variable, this is the factor (treatment) theresearcher wants to test, i.e. the cause (the condition or situation) that is altered inthe experiment by the researcher. 101
Appendix 260
Glossary
inductive reasoning also known as induction, is a kind of reasoning that constructs orevaluates general propositions that are derived from specific examples. Inductivereasoning is used to try to discover new information. whilst commonly used inscience, is not logically valid, because it is not strictly accurate to assume that ageneral principle is correct . 69, 149
information hiding It is the ability to prevent certain aspects of a software componentfrom being accessible to its clients, using for instance programming language fea-tures. The principle ensures a stable interface that protects the clients from changesin the provider’s implementation [Par72]. 12
kurtosis is a measure of the peakedness of the variable’s distribution. Higher kurtosismeans that the variance in the sample is due to infrequent extreme deviations. Alower kurtosis corresponds to frequent extreme deviations. A distribution with ahigh peak is called leptokurtic (kurtosis > 0), a flat-topped curve is called platykurtic(kurtosis < 0), and the normal distribution is called mesokurtic (kurtosis = 0). 162
liveness A Petri net with some initial marking is live if every transition is firable/reach-able from the initial marking [MTJ+10]. 33, 75, 79
non-equivalent group occurs when the researcher deliberately wants to work with dif-ferent groups, sometimes because it is impractical to select groups randomly. Theresults of the treatment of the groups are expected to be different [Tro06]. 159
non-local semantics The execution behavior of a particular node within a model maydepend on the state of other parts of the model, arbitrarily far away [Kin03]. 31
observation is some measurement that is recorded. This can be a simple activity, such asmeasuring somebody’s height or it can be the administration of a more complexinstrument such as a whole battery of questions or a coherent test. Subscripts maybe used to differentiate between observations, for example O1, O32, etc [Tro06]. 159
p-value conveys the probability that the randomness in sampling would lead to differ-ence in sample means as large as observed, even if the populations have the samemeans. It is a measure of how much evidence we have against the null hypothe-sis, which is the hypothesis of no change or no difference. The smaller the p-value,the more evidence we have against the null hypothesis. The p-value should not beinterpreted as the probability that the null hypothesis is true. An hypothesis is nota random event that can have a probability. We do not predict the happening of ahypothesis. Rather, we try to infer whether it is true or not. 162
precision is the fraction of retrieved instances that are relevant. 216
Appendix 261
Glossary
qualitative research design is scientific technique with researchers qualitatively observ-ing a domain them and trying to come up with answers which explained what theysaw. 69
random assignment is the classic method by which subjects are assigned to groups ran-domly. This should lead to similar results being achieved from each group if thesame treatment is applied. A typical experiment has two randomly assigned groups:the treatment group and the control group. When the results are compared acrossgroups, then differences should be due to the treatment. [Tro06]. 159
raw data is data not prepared to be analyzed. 165
reachability The state of a Petri net is described by the contents of the different places(marking). A marking Mn is reachable from an initial marking M0 if there exist anumber of transition firings that can lead to Mn from M0 [MTJ+10]. 33
recall is the fraction of relevant instances that are retrieved. 216
relational research is a study that investigates the connection between two or more vari-ables. The variables that are compared, are generally already present in the groupor population [Tro06]. 159
reliability yielding the same or compatible results in different experiments or statisticaltrials [Shu09]. 148, 149
resentful demoralization occurs when subjects from a group not receiving a new treat-ment feel penalized by the situation. So, they become less involved in the experi-ments compared with their counterparts. This situation is the opposite of compen-satory rivalry. 196
reversibility and home state A Petri net is reversible if for each marking M that is reach-able from some initial marking M0, there exists a finite number of transitions thatwould take the net from M back to M0 (or some other state, referred to as the homestate) [MTJ+10]. 33
sampling frame is the listing of the accessible population from which the researcherdraws the actual sample [Tro06]. 158
separation of concerns A general accepted principle in computer science that became arule of thumb for software development. The principle advocates that complexityof computer systems could be tackled by focusing the aim of specific part of the sys-tem dealing upon a single aspect. Using this principle, response to change wouldbe facilitated and a mechanism for system’s flexibility would be intrinsically partof the architecture of computer applications, provided that subsystem’s interfaceremained stable [Dij82]. 12
Appendix 262
Glossary
significance level is a value that specifies how much of the difference between the as-sumed value in the null hypothesis and the value observed from experiment is bigenough to reject the possibility that the result was a purely chance process and re-ject the null hypothesis. Common levels used in statistical analysis are 5% and 1%.162, 263
significance test is related to statistical hypothesis testing and is used to determine whetherthere is enough evidence to reject the null hypothesis. A significance test is alwaysaccompanied by a value of significance level. 172
skewness is a measure of the degree of asymmetry of a distribution. If the left tail ismore pronounced than the right tail, the function is said to have negative skewness.If the reverse is true, it has positive skewness. If the two are equal, i.e. the cumulativedensity function is symmetrical, it has zero skewness. 162
statistical significance is the minimum level at which the null hypothesis can be rejected.Lower the significance level, higher the confidence. 172
test case is a compound of test programs and test data which are expected to producecertain expected results. The purpose of the test case is the verification of the ac-complishment of some predefined system requirements, which can generate failwarnings or pass confirmations. 132
test scenario is a set of test cases that ensure that the automated process is tested fromend to end. They may be independent tests or a series of tests that follow eachother, each dependent on the output of the previous one. Sometimes, the terms testscenario and test case are used interchangeably. 134
test suite is the collection of test scenarios and/or test cases that are related or that coop-erate with each other. 132
testability whenever a hypothesis is created to prove a part of a theory, it must betestable and analyzable with current technology [Shu09]. 149
testbed is an environment that is created for testing purposes. The concept of testbeddescribe a development environment that is protected from the jeopardies of usinga production environment for testing purposes. By setting up of a testbed platform,it is possible the simulation of scenarios under conditions pretty close to the onefound in actual working environments, to rigorously test theories, tools, and tech-nologies, in a reproducible fashion. The notion of testbed can be generalized as aprocedure for testing some kinds of artifacts (e.g. model) in an isolated manner.The testbed can function as a proof of concept, by which a new artifact is testedapart from the whole where it will be later appended. 134
Appendix 263
Glossary
theoretical population is the group to which the researcher wishes to generalize thefindings of the empirical study [Tro06]. 158
trap A trap is a set of places such that every transition which inputs from one of theseplaces also outputs to one of these places. This means that once any of the placesin a trap has a token, there will always be a token in one of the places of the trap.Firing transitions may move the token between places but cannot remove a tokenfrom the trap [Pet81]. 33
treatment is an action or intervention taken that change the situation in some way. Thesecan range from a simple action such as giving the subject information to complexactivities that may range from a whole set of actions to surgical operations. Sub-scripts may be used to differentiate between treatments, for example X1, X32, etc. Ano-treatment control group may be identified with a particular notation, such as X0
or X-. Where X- is used to indicate the control group, X+ may be used to indicatethe treatment group [Tro06]. 159
validity validity encompasses the entire experimental concept and establishes whetherthe results obtained meet all of the requirements of the scientific research method.Internal validity dictates how an experimental design is structured. External valid-ity is the process of examining the results and questioning whether there are anyother possible causal relationships [Shu09]. 148
Appendix 264
AProcess Modeling Languages
A.1 Other Process Modeling Languages
• IDEF3 – http://www.idef.com/IDEF3.htm• Colored Petri Nets – http://cpntools.org/• Workflow Nets – http://woped.ba-karlsruhe.de/index.php?id=7• Role Activity Diagrams –http://www.eis.mdx.ac.uk/staffpages/geetha/BIS2000/RADs/rad.html
• Resource-Event-Agent – http://reatechnology.com/what-is-rea.html• Business Process Modeling Language –http://www.omg.org/bpmn/Documents/BPML-2003.pdf
• Business Process Definition Metamodel – http://www.omg.org/spec/BPDM/• Proposed Interchange Formats (PIF) – http://ccs.mit.edu/pif7.html• Process Specification Language (PSL) – http://www.mel.nist.gov/psl/• RosettaNet – http://www.rosettanet.org/• ebXML – http://www.ebxml.org/• BPEL4WS – http://www.ebpml.org/bpel4ws.htm• EDOC – http://www.omg.org/spec/EDOC/• π-calculus [Mil99]
CSP is a formal language for describing patterns of interaction in concurrent systems[Hoa78, BHR84]. It is a member of the family of mathematical theories of concurrencyknown as process algebras, or process calculi [Hoa04]. Industrial application of CSP tosoftware design has usually focused on dependable and safety-critical systems, namelyto analyze models to confirm that their design is free of deadlock and livelock.
CSP allows the description of systems in terms of component processes that operateindependently, and interact with each other solely through message-passing communica-tion. The relationships between different processes, and the way each process communi-cates with its environment, are described using various process algebraic operators. Usingthis algebraic approach, complex process can be easily constructed from a few primitiveelements. There are two classes of primitives in its process algebra: events that representcommunications or interactions, and primitive processes which represent fundamental be-haviors. The syntax of the language of CSP according [WG11a] is presented next.
P, Q ::= P ||| Q | P |[A]| Q | P |[ A|B ]| Q | P \ A | P M Q | P Q | P u Q | P ., Q |
e −→ P | Skip | Stop
e ::= x | x.e
– Process P ||| Q (interleaving) refers the interleaved parallel composition of processes Pand Q.
– Process P |[A]| Q (interface parallel) refers the partial interleaving of processes P and Qsharing events in set A.
– Process P |[ A|B ]| Q refers parallel composition, in which P and Q can evolve indepen-dently but must synchronize on every event in the set A ∩ B; the set A is the alphabetof P and the set B is the alphabet of Q, and no event in A ∪ B can occur without thecooperation of P and Q respectively.||| i : I • P(i) refer an indexed interleaving, |[A]| i : I • P(i) refer an partial interleaving,and || i : I • A(i) P(i) to refer parallel combination of processes P(i) for i ranging overI.
– Process P \ A (hiding) is obtained by hiding all occurrences of events in set A from theenvironment of P.
– Process P M Q (interrupt) refers a process initially behaving as P, but which may beinterrupted by Q.
– Process P Q (external choice) refers the external choice between processes P and Q;the process is ready to behave as either P or Q. An external choice over a set of indexedprocesses is written i : I • P(i).
– Process P u Q (nondeterministic choice) refers the internal choice between processes P orQ, ready to behave as at least one of P and Q but not necessarily offer either of them.Similarly an internal choice over a set of indexed processes is written u i : I • P(i).
Appendix 266
A.2. Formalizations of BPMN Verification
– Process P ., Q (sequential composition) refers a process ready to behave as P; after P has
successfully terminated, the process is ready to behave as Q.– Process e −→ P (prefixing) refers a process capable of performing event e, after which it
will behave like process P.– The process Stop is a deadlocked process and– the process Skip is a successful termination.
A.2.2 Petri Nets
Petri Nets (P/N) [Pet62, DR98] was already introduced in section 2.5.2.4. As it wasmentioned before, P/N have been extensively applied to the study of workflow [VdA94,VdAVH96]. We present next the formal definition of Petri nets according to [VdA97].
Definition A.1. Petri Net. A P/N is a triple N = (P,T,F) where:P is a finite set of places;T is a finite set of transitions (with P ∩ T = ∅);F ⊆ (P × T) ∪ (T × P) is a set of arcs (flow relation).
Place p is called an input place of transition t if and only if there exists a directed arc(p, t) ∈ F from p to t. Place p is called an output place of transition t if and only if thereexists a directed arc from t to p. The set •t refers the set of input places for transition t.The set t• refers the set of output places for transition t. The sets •p and p• hence haveequivalent meanings concerning place p. A P/N is strongly connected if and only if, forevery pair of nodes x, y ∈ P ∪ T, there is a directed path leading from x to y.
During the execution of a P/N, each place holds zero or more tokens, and a state ofa P/N, called a marking, is a function from each place in the net to a number of tokensit holds. In a given marking, a transition t is enabled if every input place in •t has atleast one token; firing t removes a token from every input place in •t and adds a token toevery output place in t•. The state of the P/N transitions from one to the next by firingany one of the enabled transitions. A marking of a P/N is dead if it does not enable anytransition, and a transition in a P/N is dead if and only if the net has no marking thatenables it.
A.2.3 Web Ontology Language
Web Ontology Language (OWL) is a family of knowledge representation languages forexpressing ontologies. The languages are characterized by formal semantics and RDFserializations for the Semantic Web [MPSP+09]. The Semantic Web provides a commonframework that allows data to be shared and reused across application, enterprise, andcommunity boundaries. RDF is closely related to semantic networks, since it is a graph-based data model with labeled nodes and directed, labeled edges. This is a flexible modelfor representing data. The fundamental unit of RDF is the statement, which correspondsto an edge in the graph. An RDF statement has three components: a subject, a predicate,
Appendix 267
A.2. Formalizations of BPMN Verification
and an object. The subject is the source of the edge and must be a resource. In RDF, aresource can be anything that is uniquely identifiable via a Uniform Resource Identifier(URI). OWL has three variants with different levels of expressiveness: OWL Lite, OWLDL and OWL Full. Each of these sub-languages is a syntactic extension of its simplerpredecessor. As an ontology language, OWL is primarily concerned with defining termi-nology that can be used in RDF documents, i.e., classes and properties. Most ontologylanguages have some mechanism for specifying a taxonomy of the classes. In OWL, youcan specify taxonomies for both classes and properties.
Semantically, OWL is based on Description Logics (DL). DL are a family of logicsthat are decidable fragments of first-order predicate logic. These logics focus on describ-ing classes and roles, and have a set-theoretic semantics. Different description logics in-clude different subsets of logical operators [Hef04]. Two of OWL’s sub-languages closelycorrespond to known description logics: OWL Lite corresponds to the description logicSHIF(D) and OWL DL corresponds to the description logic SHOIN(D) [HPSVH03].
A common property of ontologies and the OWL semantics is the so-called open-worldassumption [HPSVH03], a form of partial description or under-specification as a meansof abstraction, i.e., from the absence of statements, a deductive reasoner must not inferthat the statement is false.
In [Nat11] it is defined an ontology that formally represents the BPMN specification.(BPMN 2.0 Ontology) and can be used as a knowledge base. The description of an ele-ment is combined within the corresponding class and further explanations are providedin annotations. This is claimed to allow a much faster understanding of BPMN. In addi-tion, the ontology is used as a syntax checker to validate concrete BPMN process models.
A.2.4 Abstract State Machines
Unlike natural language, the Abstract State Machines (ASM) method has precise seman-tics, i.e., there’s never any doubt about the meaning of an ASM. Abstraction is anotherimportant benefit of the ASM method. ASMs can be executed directly using varioustools; this makes the transitions from designing to coding, and from designing to testing,much easier [HW02]. Based closely in [SSB01], we present next the main foundations ofASM.
An ASM is a system of finitely many transition rules of form
if Condition then Updates
which transform abstract states. The Condition (so called guard) under which a rule isapplied is an arbitrary first-order formula without free variables. Updates is a finite set offunction updates (containing only variable free terms) of form
f(t1, . . . , tn) := t
whose execution is to be understood as changing (or defining, if there was none) thevalue of the (location represented by the) function f at the given parameters.
Appendix 268
A.2. Formalizations of BPMN Verification
The notion of ASM states is the classical notion of mathematical structures where datacome as abstract objects, i.e., as elements of sets (domains, universes, one for each categoryof data) which are equipped with basic operations (partial functions) and predicates (at-tributes or relations). Without loss of generality one can treat predicates as characteristicfunctions.
The notion of ASM run is the classical notion of computation of transition systems. AnASM computation step in a given state consists in executing simultaneously all updates ofall transition rules whose guard is true in the state, if these updates are consistent. For theevaluation of terms and formulae in an ASM state, the standard interpretation of functionsymbols by the corresponding functions in that state is used.
Simultaneous execution provides a convenient way to abstract from irrelevant se-quentiality and to make use of synchronous parallelism. This mechanism is enhanced bythe following concise notation for the simultaneous execution of an ASM rule R for eachx satisfying a given condition ϕ:
forall x with ϕ do RA priori no restriction is imposed neither on the abstraction level nor on the complex-
ity nor on the means of definition of the functions used to compute the arguments ti andthe new value t in function updates. The major distinction made in this connection fora given ASM M is between static functions – which never change during any run of M– and dynamic ones which typically do change as a consequence of updates by M or bythe environment (i.e., by some other agent than M). The dynamic functions are furtherdivided into four subclasses.• Controlled functions (for M) are dynamic functions which are directly updatable by
and only by the rules of M, i.e., functions f which appear in a rule of M as leftmostfunction (namely in an update f(s) := t for some s,t) and are not updatable by theenvironment.• Monitored functions are dynamic functions which are directly updatable by and
only by the environment, i.e., which are updatable but do not appear as leftmostfunction in updates of M.• Interaction functions are dynamic functions which are directly updatable by rules of
M and by the environment.• Derived functions are dynamic functions which are not directly updatable neither
by M nor by the environment but are nevertheless dynamic because defined (forexample by an explicit or by an inductive definition) in terms of static and dynamicfunctions.
Appendix 269
[This page is intentionally blank]
BA Catalog of BPMN Patterns and
Anti-Patterns (Sample)
B.1 A Top-Level Process can only be instantiated by a restrictedset of Start Events types
Any container (process or subProcess) that does not have a parent container is considereda top-level Process [BPM11, page 238]. Top-level processes can have one of seven typesof start events (see Figure 3.3, first column): none, message, timer, conditional, signal,multiple, and parallel [BPM11, page 112].
271
B.2. Outgoing Sequence Flow not allowed in an End Event.
Figure B.1: A Top-Level Process can only be instantiated by a restricted set of Start Eventstypes
Correct: A top-level process is instantiated by allowed types of start events (none and amessage types) (top)
Wrong: A top-level process is instantiated by an invalid type of start event (escalationtype) (bottom)
B.2 Outgoing Sequence Flow not allowed in an End Event
The end event indicates where a process will end. In terms of sequence flows, the endevent ends the flow of the process, and thus, will not have any outgoing sequence flows[BPM11, page 249].
Appendix 272
B.3. Outgoing Message Flow not allowed in a Catch Event.
Figure B.2: Outgoing Sequence Flow not allowed in an End Event
Correct: End Event has no outgoing sequence flows (top)
Wrong: End Event has an outgoing sequence flow (bottom)
B.3 Outgoing Message Flow not allowed in a Catch Event
A Start Event or a catching Intermediate Event cannot have outgoing message flows[BPM11, page 251].
Figure B.3: Outgoing Message Flow not allowed in a Catch Event
Correct: Catch Events with incoming Message Flow (top)
Wrong: Catch Events with outgoing Message Flow (bottom)
Appendix 273
B.4. A Catch Event with incoming Message Flow must have Message or Multiple type
B.4 A Catch Event with incoming Message Flow must have Mes-sage or Multiple type
A Start Event or a catching Intermediate Event with incoming Message Flow must be oftype Message of Multiple [BPM11, pages 44, 271].
Figure B.4: A Catch Event with incoming Message Flow must have Message or Multipletype
Correct: Catch Events with incoming Message Flow are Message or Multiple types (top)
Wrong: A Start Event with incoming Message Flow cannot be of Timer type (middle) oruntyped (bottom)
Appendix 274
B.5. Explicit Start/End Events do not allow Activities or Gateways withoutincoming/outgoing Sequence Flow
B.5 Explicit Start/End Events do not allow Activities or Gate-ways without incoming/outgoing Sequence Flow
Start Event and End event are optional [BPM11, page 238]. However, if there is at leastone explicit Start/End Event in a container (Process or SubProcess), there must not beother flow nodes such as Activity and Gateway, without incoming/outgoing sequenceflow [BPM11, pages 153, 289, 430]. There are some exceptions: Compensation Activityan Event SubProcess do not have incoming and outgoing Sequence Flows.
Figure B.5: Explicit Start/End Events do not allow Activities or Gateways without in-coming/outgoing Sequence Flow
Correct: A Start event and activities or gateways without incoming sequence flow
Wrong: A Start event and an activity without incoming sequence flow
Appendix 275
B.6. A conditional Sequence Flow cannot be used if there is only one sequence flow outof the element
B.6 A conditional Sequence Flow cannot be used if there is onlyone sequence flow out of the element
Figure B.6: A conditional Sequence Flow cannot be used if there is only one sequenceflow out of the element
Correct: The sequence flow from Activity1 to Activity2 does not have any decoratorsince it is unique (top)
Wrong: The sequence flow from Activity1 to Activity2 has a condition (bottom)
Appendix 276
B.7. A Boundary Event must have exactly one outgoing Sequence Flow (unless it hasthe Compensation type)
B.7 A Boundary Event must have exactly one outgoing SequenceFlow (unless it has the Compensation type)
A Boundary Event is attached to an Activity and an outgoing exception flow comes outfrom it, through a Sequence Flow. Exactly one Sequence Flow is allowed from a Bound-ary Event except in the case that it is of type Compensation. In this particular case anAssociation can replace or not the Sequence Flow [BPM11, pages 259, 440, 441].
Figure B.7: A Boundary Event must have exactly one outgoing Sequence Flow (unless ithas the Compensation type)
Correct: Only one sequence flow has as source a Boundary Event (top)
Wrong: More than one sequence flow has as source a Boundary Event (bottom)
Appendix 277
B.8. Use a Timer Intermediate Event with an Event Gateway
B.8 Use a Timer Intermediate Event with an Event Gateway
One way for a modeler to ensure that the Process does not get stuck at an Event BasedExclusive Gateway is to use a Timer Intermediate Event as one of the options for theGateway [WM08].
Figure B.8: Use a Timer Intermediate Event with an Event Gateway
Correct: A Timer Intermediate Event in a set including an Event Gateway (top)
Wrong: An Event Gateway without a Timer Intermediate Event(bottom)
Appendix 278
B.9. Use a Default Condition at an Exclusive Gateway
B.9 Use a Default Condition at an Exclusive Gateway
One way for the modeler to ensure that the Process does not get stuck at an ExclusiveGateway is to use a default condition for one of the outgoing Sequence Flow. This createsa Default Sequence Flow. The Default is chosen if all the other Sequence Flow conditionsturn out to be false [WM08].
Figure B.9: Always use a Default Condition with an Exclusive Gateway
Correct: Use of a Default Condition with an Exclusive Gateway (top)
Wrong: An Exclusive Gateway without Default Condition (bottom)
Appendix 279
B.10. Two Activities in the same Process should not have the same name
B.10 Two Activities in the same Process should not have the samename
It is highly recommended that the activities’ names be unique. If it is required an activityto be reused in a process, a Global Activity should be used instead [Sil09].
Figure B.10: Activities on the same Process should have different names
Correct: Two Activities in the same Process with different names but invoking the sameGlobal Activity (top)
Wrong: Two Activities in the same Process with the same name (bottom)
Appendix 280
CData Collection for a Survey on
Effectiveness of Current BPMN Toolson Detection of Rules Violations on
Process Models
Figure C.1: Model-snippet implemented in Adonis Community Edition (Version:2.01.00.812)
281
Figure C.2: Model-snippet implemented in Aris Express (Version: 2.4)
Figure C.3: Model-snippet implemented in Bizagi (Version: 2.3.0.5)
Figure C.4: Model-snippet implemented in Enterprise Architect (Version: 9.0.908)
Appendix 282
Figure C.5: Model-snippet implemented in eClarus (Version: 2.1.0.200904272037)
Figure C.6: Model-snippet implemented in iGrafx Process 2013 (Version: 15.0.1.1547)
Figure C.7: Model-snippet implemented in MagicDraw (Version: 17.0.3)
Appendix 283
Figure C.8: Model-snippet implemented in Modelio (Version: 2.2.1)
Figure C.9: Model-snippet implemented in Signavio (Version: 6.2)
Appendix 284
Figure C.10: Model-snippet implemented in TIBCO (Version: 3.5.3.022)
1.2 Accounts payable (AP) can be an area of fi-nance where tremendous inefficiencies can oc-cur. All companies are looking to: Reduce costsand eliminate errors with data entry; Maximizesettlement discounts; Minimize late paymentpenalties; Eliminate lost invoices; Improve cashflow; Foster healthy vendor relationships.
Trisotech Acquisitionfollowing RFQ
1.2 Acquisition process following Request for Quo-tation from selected suppliers (BPMN v1.2Specification)
Trisotech Airline Check In 1.2 Activities performed by a Airline agent respon-sible for checking in an airline passenger andits baggages.
Trisotech Bank AccountOpening
1.2 The sequence and the wording of few activitiesof banking organizations.
Trisotech Book Writing andPublishing
1.2 This process presents a way to synchronizehighly independent activities occuring duringthe writing of a book to be published. Pro-cess used in Stephen A. White and Derek Miersbook "BPMN Modeling and Reference Guide".
Trisotech BPI Web Reg-istration withModerator
1.2 This process illustates the activities performedby a user and by Business Process Incubator(BPI) to handle a BPI membership request. Thisversion includes the presence of a "Moderator"to review and approve teh registration.
Trisotech BPI Web Regis-tration withoutModerator
1.2 This process illustates the activities performedby a user and by Business Process Incubator(BPI) to handle a BPI membership request.
Trisotech Change of Ad-dress
1.2 Each year, 14% of the U.S. population will moveresulting in over 120,000 change of address(COA) requests that need to be processed atBigger Bank.
Trisotech Claims Process-ing
1.2 Insurance company explored alternatives toimprove an existing process and reduce oper-ational costs.
Appendix 288
BPDs used in Empirical Validation
Source Name BPMN Description
Trisotech CommercialFinancing
1.2 Process activities performed by the financial in-stitution personnel involved in the sales, anal-ysis and approval of Commercial Loans. Thisprocess span the activities from the reception ofa loan request to to dipursement of the fonds tothe approved requester.
Trisotech Credit Reviewand Approval
1.2 High Level Credit Review Process with Startand End Events with and withou triggers, al-lowing this process to be used as a standaloneprocess or a Reusable sub-process. This dia-gram correponds to the Figure 9.12 (page 61) ofBPMN v1.2 Specifications.
Trisotech Customer QuoteRequest
1.2 A Customer makes a request for a quote froma supplier befor confirming or infirming an for-mal order.
Trisotech Email Voting 1.2 It is a process for resolving issues through e-mail votes. The email voting presented in theBPMN v1.2 Specifications.
1.2 This process presents the activities required toaccept, analyse, approve and pay an expensestatement submitted by an employee. This pro-posed alternative assume that email sending toemployees does not create process delays.
1.2 This process presents the activities required toaccept, analyse, approve and pay an expensestatement submitted by an employee. Thisalternative remove the assumption that emailsending to employees does not create processdelays.
Trisotech HR EmployeeOn-boarding #1
1.2 HR needs to meet 45-day SLA to complete theon-boarding of an employee.
Trisotech HR EmployeeOn-boarding #2
1.2 With 300 - 400 employees joining and departingthe company every two months, the HR orga-nization is constantly recruiting, on-boarding,training, and relieving staff. Achieving SLAswas a challenge.
Trisotech Incident Manage-ment as DetailedCollaboration
2.0 This process illustates the ping-pong-game ofaccount manager, support agents and softwaredeveloper by switching from a single-pool-model to a collaboration diagram.
Trisotech Insurance ClaimProcessing
1.2 This process presents the high level activitiesused to handle an Insurance Claim. Only theInsurance company Private Process is shown inthis example.
Trisotech LaserTec Produc-tion Process
1.2 Present an end to end process of the LaserTecproduction. Used by Alexander Grosskopf,Gero Decker and Mathias Weske in their book"The Process - Business Process Modeling usingBPMN"Appendix 289
BPDs used in Empirical Validation
Source Name BPMN Description
Trisotech Loan Processing 1.2 The bank used process simulation to assess theimpact of upcoming marketing programs thatwere aimed at first-time auto buyers.
Trisotech Mortgage Ap-proval
1.2 This process illustrates the usage of SignalEvents to coordinate the decisions of a Cus-tomer Service Representative and a Managerworking in a Mortgage Approval Business Pro-cess. Process used in Stephen A. White andDerek Miers book "BPMN Modeling and Ref-erence Guide".
Trisotech New Car Sales 1.2 Present in the book "BPMN Method & Style"from Bruce Silver
Trisotech Order Fulfillment 2.0 Present in the book "BPMN Method & Style"from Bruce Silver
Trisotech Order fulfillment- BPMN v2.0Specifications
2.0 In this example, following an order, parts areretrieve (if available) and/or procured (if notavailable). If all parts are made available ontime, the order is accepted, otherwise it is re-jected. In both cases the customer is notified.
Trisotech Order Fulfillmentand Procurement
2.0 This order fulfillment process starts after re-ceiving an order message and continues tocheck whether the ordered article is available ornot. An available article is shipped to the cus-tomer followed by a financial settlement, whichis a collapsed sub-process in this diagram. Incase that an article is not available, it has tobe procured by calling the procurement sub-process.
Trisotech Order Processing 1.2 High level process presenting exclusively thenormal flow of an order reception and fulfill-ment. This BPMN diagram is from the BPMNv1.2 Specifications
Trisotech Order Processingwith Credit CardAuthorization
1.2 Simple Collaboration process requiring CreditCard Authorization from a Financial Institutionbefore a Supplier process an order and ship it.It is presented in the BPMN v1.2 Specifications
Trisotech Patient Treatment- Abstract Process
1.2 Patient Illness Treatment Process used as exam-ple in the BPMN 1.2 Specifications. This dia-gram presents the Doctor’s Office Abstract Pro-cess, corresponding to Figure 7.2 (page 13) ofBPMN v1.2 Specifications.
Trisotech Patient Treatment- Collaboration
1.2 Patient Illness Treatment Process used as exam-ple in the BPMN 1.2 Specifications. This dia-gram presents the Collaboration Business Pro-cess, showing Patient and Doctor Office activ-ities and the message flows between them. Itcorresponds to Figure 7.3 (page 14) of BPMNv1.2 Specifications.
Appendix 290
BPDs used in Empirical Validation
Source Name BPMN Description
Trisotech Pizza Co. Deliv-ery Process
1.2 This process is presented at a very high level inPaul Harmon book "Business Process Change -A Guide for Business Managers and BPM andSix Sigma Professionals". It has been detailedby Trisotech to support its BPM - Biik of Knowl-edge material.
Trisotech Property and Ca-sualty InsuranceClaim Processing
1.2 This process outlines property and casualtyinsurance claim processing from the point ofview of the insurance company. This MSProject Template was designed for automo-bile insurance, but can also be used for home-owner’s insurance.
Trisotech Shipment Processof a HardwareRetailer
2.0 This process has only one pool and differentlanes for the people involved in this process,which automatically means that it is blankedout the communication between those people:It is just assumed that they are communicatingwith each other somehow. If we had a processengine driving this process, that engine wouldassign user tasks and therefore be responsiblefor the communication between those people.
Trisotech Stock mainte-nance Process
2.0 This process deals with the case that an articleis not available, it has to be procured by callingthe procurement sub-process.
Trisotech The Nobel Prize 2.0 The processe is slightly differ for each of the sixprizes; the results are the same for each of thesix categories. Following is the description forthe Nobel Prize in Medicine.
Trisotech The Pizza Collab-oration
2.0 This example is about Business-To-Business-Collaboration. It intends to model the interac-tion between a pizza customer and the vendorexplicitly.
Trisotech Travel Booking 1.2 The Travel booking process illustrates the us-age of a transaction Sub-process. The process ispresented in Bruce Silver book "BPMN Methodand Style"
Trisotech Travel Book-ing with EventSub-processes
2.0 The Travel booking process illustrates the us-age of a transaction Sub-process. In this ex-ample flights and hotel reservation are made,however they can be reversed if some unde-sirable results or events occur. The process ispresented in Bruce Silver book "BPMN Methodand Style"
Appendix 291
BPDs used in Empirical Validation
Source Name BPMN Description
Bizagi 20-F SOX Man-agement process
1.2 Bizagi’s 20-F SOX Management template facil-itates the preparation of the Form 20-F neededto comply with the Sarbanes Oxly Act. It cov-ers the definition of tasks and responsibilitieswith the visibility, availability and traceabilityneeded by auditors, ensuring transparency andcontrol of activities.
Bizagi Access Manage-ment process
1.2 This template is based on the principals of ITILpractices to help guarantee the availability ofinformation to the users that really need it.This template includes the creation of requestsfor activating or deactivating permissions overapplications, modules, folders or services, aswell as managing approvals and permissionupdates.
Bizagi accounts payableprocess
1.2 The Accounts Payable process covers the var-ious tasks involved in the invoices reception,their validation and approval, reducing processtime and avoiding incorrect information.
Bizagi Ad-Hoc Process 1.2 The Ad Hoc process template is a workflowpattern that allows handling unstructured pro-cesses rather than well predefined processes.With the Ad Hoc pattern you can create tasksat any time that can be allocated to anyone andcan be performed in any order during the life-time of a process.
Bizagi Change Manage-ment process
1.2 Change Management is based on the princi-pals of ITIL V3 to guarantee a proper imple-mentation of technological changes. It allowsa complete planning, risk and impact assess-ment, and an appropriate communication, im-plementation and documentation.
Bizagi Offboarding pro-cess
1.2 Offboarding Process helps the Human Re-sources Area to execute all activities neededfor the departure of an employee. It autom-atizes and reduces the performing time in ac-tions such as stopping payroll, benefits andsecurity access, and collecting the company’sproperty.
Bizagi Onboarding pro-cess
1.2 Onboarding process assists companies in thecoordination of the activities required in the en-try of new employees. It focuses on integratingnew employees into the organization, prepar-ing them to execute their functions properlyand to quickly become productive members ofthe organization.
Appendix 292
BPDs used in Empirical Validation
Source Name BPMN Description
Bizagi personal loans re-quest process
1.2 This process covers the various tasks involvedin the application/request for credit products,made by individuals through a bank or bank-ing entity.
1.2 This process template will help you to managePetitions, Claims, Complaints and Suggestionsthrough the definition and control of the neces-sary activities to solve them.
Bizagi purchase requestprocess
1.2 This process handles the whole purchase pro-cess: purchase request, approval (approval lim-its), quotations, supplier selection, purchase or-der, invoice control.
Bizagi Recruitment andSelection process
1.2 This template covers all the different activitiesperformed to find a person to fill a vacancy.Companies can reduce the time it takes to hirea new employee and control activities such asscheduling and collecting the result of tests, as-signing interviews, updating the list of candi-dates, etc.
Bizagi Six sigma projectmanagementprocess
1.2 This process helps to perform and manage SixSigma projects that use DMAIC methodology.The process reduces the task assignment timeand improves the manner in which informationis collected and the project is presented.
Bizagi travel requestprocess
1.2 This process handles travel requests (flight, ho-tel, cash advance) and approvals for employ-ees. At the conclusion of the travel the processalso handles the expense report.
Bizagi vacation leave re-quest process
1.2 This process handles vacation leave requestsand approvals for employees.
Bizagi vehicle insurancepolicy underwrit-ing process
1.2 This process handles the initial study of the ve-hicle, the generation of the insurance quotation,the study of the viability and risks, the inspec-tion and, finally, the issue of the insurance pol-icy itself.
Appendix 293
[This page is intentionally blank]
EWell-formedness Rules for BPMN
Metamodel
295
Table E.1: Preciseness Rules for BPMN MetamodelId. Description
1 Complement the Metamodel with Flow Control Well-formedness Rules1.1 Collaboration1.1.1 Only one implicit pool can exist in a collaboration1.2 Process1.2.1 A Process is an abstraction of more detailed Process, if its normal flow is
contained by the second Process’s normal flow1.2.2 A public Process may not be executable1.2.3 A Top-Level Process can only be instantiated by a restricted set of Start Events
types1.2.4 A Call Activity must only call a Callable Element1.3 Sub-Process1.3.1 The messages flows interacting with a collapsed view of a sub-process must
be the same as the exchanged in the detailed view of the same sub-process1.3.2 A Sub-Process can only have one None Start Event1.3.3 An Event Sub-Process can only have a single Start Event and it must be typed1.3.4 An Event Sub-Process must not have any incoming or outgoing Sequence
Flows1.4 Flow Nodes1.4.1 If a container includes Start and End Events, as general rule, all Flow Nodes
must have at least one incoming or one outgoing Sequence Flow1.4.2 A Flow Node, in a container that includes start and end events, must have at
least one incoming or one outgoing sequence flow.1.5 Events1.5.1 Only some predefined types of Start, Intermediate and End Events are al-
lowed in specific contexts1.5.2 Incoming Sequence Flow not allowed in a Start Event.1.5.3 Outgoing Sequence Flow not allowed in an End Event.1.5.4 A Catch Event with incoming Message Flow must have Message or Multiple
type1.5.5 A Catch Event must have Multiple type if there are more than one incoming
Message Flow1.5.6 Outgoing Message Flow not allowed in a Catch Event.1.5.7 A Throw Event with outgoing Message Flow must have Message or Multiple
types.1.5.8 A Throw Event must have Multiple type if there are more than one outgoing
Message Flow1.5.9 Incoming Message Flow not allowed in a Throw Event.1.5.10 A Catch Link Event must have an incoming Sequence Flow. A Throw Link
Event must have outgoing Sequence Flow
Appendix 296
Preciseness Rules for BPMN Metamodel
Id. Description
1.5.11 Multiple Throw Link Events allowed, but only one Catch Link Event1.5.12 Source and target Link Events must have names.1.5.13 Throw and Catch Link Events names must match in the same container1.5.14 Intermediate Events used within normal flow require incoming and outgoing
Sequence Flows1.5.15 Explicit Start/End Events do not allow Activities or Gateways without in-
coming/outgoing Sequence Flow1.5.16 Activities or Gateways without incoming Sequence Flow do not allow ex-
plicit Start Events1.5.17 Implicit Start Events require implicit End Events, and vice versa1.5.18 Non-interrupting Start Events are only allowed in Event SubProcess1.5.19 Error intermediate events can only be attached to activity boundaries1.5.20 Catch Error Event must trigger an exception flow1.5.21 A Throw Error Event must have an unnamed Catch Error Event or a Catch
Error Event with the same name1.5.22 A catching Error Event name must match the name of a thrown Error Event
or be unnamed1.5.23 Unnamed and named Catch Error Events must not be mixed1.5.24 A Throwing Error Event must be an End Event1.5.25 Catch Escalation Events can only be attached to activity boundaries1.5.26 Catch Escalation Event must trigger an exception flow1.5.27 A Throw Escalation Event must have an unnamed Catch Escalation Event or
a Catch Escalation Event with the same name1.5.28 Named and unnamed Interrupting Catch Escalation Events must not be
mixed1.5.29 Unnamed and named Catch non-Interrupting Escalation Events must not be
mixed1.5.30 A Throwing Escalation End Event must be caught by an Interrupting Escala-
tion Catch Event1.5.31 A Throwing Escalation Intermediate Event must be caught by an non-
Interrupting Escalation Catch Event1.5.32 A Throw Escalation Event must have an unnamed Catch Escalation Event or
a Catch Escalation Event with the same name1.5.33 A Throw Signal Event must be caught by a Catch Signal Event with the same
name or unnamed.1.5.34 A named Catch Signal Event captures Signal Throw Event with the same
name.1.5.35 A catching Cancel Intermediate Event can only be used attached to the
boundary of a Transaction Sub-Process.1.5.36 The Cancel End Event must be contained within the Transaction Sub-Process
or within a lower-level child Transaction Sub-Process1.5.37 A Terminate End Event must exist in a Transaction if there are several types
of End Events1.5.38 A Compensation End or Intermediate Event can only be used in a Sub-
Process which is not a Transaction1.5.39 Embedded Sub-Process can have Compensation Activities explicitly called or
via Event Sub-Process1.5.40 The name of the throwing Intermediate Compensation Event must match to
the name of canceled Activities.Appendix 297
Preciseness Rules for BPMN Metamodel
Id. Description
1.5.41 An exception flow originated by Interrupting Catch Events can merge thenormal flow through an Exclusive Gateway
1.5.42 An exception flow originated by Non-Interrupting Catch Events can mergethe normal flow through an Inclusive Gateway
1.5.43 A condition Expression must not be used if the Source of the Sequence Flowis an Event
1.5.44 A Boundary Event must have exactly one outgoing Sequence Flow (unless ithas the Compensation type)
1.5.45 A Boundary Event must not have incoming Sequence Flow1.5.46 An Intermediate Event must have at least one incoming and outgoing Se-
quence Flow1.6 Gateway1.6.1 A Parallel Gateway joins only non-exclusive Sequence Flows1.6.2 A join Exclusive Gateway must merge only exclusive Sequence Flows1.6.3 A Data-Based Exclusive Gateway must have exclusive outgoing Sequence
Flows1.6.4 A Gateway must have either multiple incoming Sequence Flow or multiple
outgoing Sequence Flow (i.e., it must merge or split the flow).1.6.5 A Gateway with a gatewayDirection of converging must have multiple in-
coming Sequence Flow, but must not have multiple outgoing Sequence Flow.1.6.6 A Gateway with a gatewayDirection of diverging must have multiple outgo-
ing Sequence Flow, but must not have multiple incoming Sequence Flow.1.6.7 An Event Gateway must have two or more outgoing Sequence Flow.1.6.8 A Conditional Sequence Flow must not be used if the source Gateway is of
type Event-Based.1.6.9 A condition Expression must be defined if the Source of the Sequence Flow
is an Exclusive or Inclusive Gateway.1.6.10 Target of the Event Based Gateway must be Receive Task or specific Interme-
diate Catch Event1.6.11 Receive Tasks used in an Event Based Gateway configuration must not have
any attached Boundary Event.1.6.12 If Message Intermediate Catch Events are used as Target for the Gateway’s
outgoing Sequence Flow, then Receive Tasks must not be used and vice versa.1.6.13 Target elements in an Event Gateway configuration must not have any addi-
tional incoming Sequence Flow (other than that from the Event Gateway).1.6.14 A Parallel or Complex Gateway must not have outgoing Conditional Se-
quence Flow.1.7 Activity1.7.1 An Activity with multiple Conditional Sequence Flows must have at least
two outgoing Sequence Flows1.7.2 A Compensation Activity must not have any incoming or outgoing Sequence
Flow1.7.3 A Compensation Activity must reside within the Process of the Activity that
will be compensated1.7.4 A Receive Task must not have an outgoing Message Flow.1.7.5 A Send Task must not have an incoming Message Flow.1.7.6 A Script or Manual Task must not have an incoming or an outgoing Message
Flow.Appendix 298
Preciseness Rules for BPMN Metamodel
Id. Description
1.8 Sequence Flow1.8.1 A conditional Sequence Flow cannot be used if there is only one sequence
flow out of the element.1.8.2 Sequence Flows cannot cross container boundaries.1.8.3 The source and target must not be the same.1.9 Message Flow1.9.1 A Message Flow can only have as source Message End or Intermediate Throw
Event; Send, User, or Service Task; Subprocess; or “black box” pool.1.9.2 A Message Flow can only go to a Message Start or Intermediate Catch Event,
Boundary Event; Receive, User, or Service Task; Subprocess; or “black box”pool.
1.9.3 A Message Flow should not connect to the border of a "white Box" Pool.1.9.4 A Message must be attached to a Message Flow or must be connected to an
Association connecting to a Message Flow, a Send Task or a Receive Task, ora Message Event Definition.
1.10 Artifacts1.10.1 An Association that is connected to a Text Annotation should have a None
Direction.1.10.2 An Association should not connect two Text Annotations.1.10.3 An Association must connect to a Text Annotation.2 Complement the Metamodel with Data Flow Well-formedness Rules2.1.1 A reusable SubProcess has only self-contained data2.1.2 A Data Object must have at least one connected Data Association.2.1.3 A Data Store must have at least one connected Data Association.2.1.4 Data Object References can only access Data Objects residing in the same
container or their parents.3 Complement the Metamodel with Modeling Best-Practices Recommendations3.1 Process3.1.1 Use only 7 ± 2 Flow Nodes per diagram3.2 Event3.2.1 Use Send/Receive Task or Throw/Catch Message Intermediate Events (not
both)3.2.2 Use explicitly Start Events and End Events3.2.3 It is recommended that only one Start Event be used.3.2.4 Use a Default Condition when using Conditional Sequence Flow3.2.5 Use a Timer Intermediate Event with an Event Gateway3.2.6 An event has at most one outgoing Sequence Flow3.2.7 A Start Event should have a name3.2.8 A Message Start Event should have an incoming Message Flow.3.2.9 A Catching Intermediate Message Event should have an incoming Message
flow.3.2.10 A Throwing Intermediate Message Event should have an outgoing Message
Flow.3.2.11 A Intermediate Event must have a name3.2.12 An end event should be labeled with the name of the end state.3.2.13 If a SubProcess is followed by a yes/no gateway, at least one end event of the
SubProcess should be labeled to match the gateway label.
Appendix 299
Preciseness Rules for BPMN Metamodel
Id. Description
3.3 Gateway3.3.1 Use a Default Condition at an Exclusive Gateway3.3.2 Use a Default Condition at an Inclusive Gateway3.3.3 Match merging and splitting Sequence Flow in Parallel Gateways (if the de-
sired behavior is to merge them again).3.3.4 Ensure that the number of incoming Sequence Flow is correct for a Parallel
Gateway3.3.5 Match merging and splitting Inclusive Gateways3.3.6 Use a Gateway as mediator when merging exclusive paths3.3.7 Simultaneous merging and splitting gateway should be avoid3.3.8 An Exclusive or Event Gateway should have at most one unnamed outgoing
Sequence Flow.3.3.9 An Inclusive Gateway should have all outgoing Sequence Flow named.3.3.10 If a SubProcess is followed by a yes/no Gateway, at least one End Event of
the SubProcess should be named to match the Gateway name.3.4 Activity3.4.1 Activities should be named3.4.2 Two Activities in the same Process should not have the same name.3.4.3 A Send Task should have an outgoing Message Flow.3.4.4 A Receive task should have an incoming message flow.3.4.5 If a SubProcess is followed by a Gateway labeled as a question, it must have
more than one End Event.3.5 Message Flow3.5.1 A Message Flow should be named with the name of the Message.
Appendix 300
FA Business Process of Financial
Services Provisioning
F.1 Introduction
This document describes the business process implemented by financial institutions forproviding financial self-services to customers, in public spaces. The business process is acompound of human and automated activities, supported by a computerized telecom-munications device known as automated teller machine (ATM).
With the implementation of this business process, the financial institutions have costsavings due to customers operating in self-service mode, i.e. without direct sup-port ofan institution’s employees, and also by benefiting from institution’s brand exhibition, aswell as the profits attained from advertising sold through ATMs.
To access the services, a customer must have an account, at one of the banks associatedto the ATM network, and use a magnetic card to interact with an ATM. Customers areable to do financial activities such as: to query their account balance, to withdraw cash(i.e., take money out of an account) and deposit funds (i.e., place cash or checks intoan account). Beside those automatic activities, human activities must also be done, inthe context of the business process to provide quality financial services to its customers.Among those activities we name a few such as cash replenishment, consumables (toner,paper) replacement, and deposit envelopes removal. Before the financial services is madeavailable to customers, a critical task is the design and deploy of the business process, andthe correspondent IT service, support-ed by the ATM.
First of all, the venue of ATMs installation must be adequately planned. ATMs spots
301
F.1. Introduction
might include, beside the own financial institutions facilities (i.e. branches), other lo-cations such as shopping centers. Making deals with store owners and gas stations forthe housing could be also an option. Busiest locations (that generate an average of 600transactions a month) are preferable, which means spots that get good foot traffic andlot of people buying things. Less visited places should have a minimum around 300transactions a month. When seeking for a location, the key is to find a place where therevenue will outweigh the installations and the ATM’s operation costs. The housing costcan amount, depending on how busy is the place, from 30e to 400e a month. Regardingrevenues, from receipt coupons to multimedia ads, automated teller machines can drawprofits, on average between 1,500e and 3,000e per ATM per month.
The business process that provides the financial service must have certain quality at-tributes (also known as non-functional requirements such as performance1, availability2).For instance, in terms of performance, the maximum acceptable response time by ATM’soperations must be 5 seconds. Regarding availability, the monetary cost of downtime, ifa machine does 600 transactions a month, could be estimated in 60e per day. Besidesthis tangible cost, there are less quantifiable costs such as the financial institution lossof customer loyalty and business reputation, since this financial service is a kind of bill-board for the bank’s brand. So, the financial service would require an availability forinstance of 99%, which would be equivalent to a non-functional requirement such as defollowing one: the service must be online every day from 00:00 to 24:00 with only onehour off for maintenance; this means that from 00:00 to 24:00 and outside the mentionedperiod of maintenance, in 365 days of the year only 0.01 of the time (83.95 hours in to-tal and no more than 2 hours each time from 08:00 to 23:00) the financial service can benon-operational.
For ensuring the above mentioned quality attributes of financial services, the busi-ness process must be carefully administered, by the service desk. This department of ITmonitors the messages sent and received regarding the financial service. Those messagesare related to its regular transactions as well as component malfunction, running out ofpaper or toner, the need for cash replenishment or deposits removal.
In case of an incident (e.g. lack of paper or toner, lack of cash, deposits containerfull), either members of the financial institution (e.g. a bank branch on regular hours),or outsourcing partners (a security services company if the incident occurred out of officehours), are warned by service desk to solve the cause of the incident, within a period ofsay 2 hours, specified in the service level agreement (SLA) contract previously signed on.
For accounting of resources usage (efficiency) the contract with the security ser-vicescompany costs 50e per visit. One month of security service costs 200e per ATM, sinceeach ATM requires a minimum of four visits per month. Therefore, each ATM needs at
1Performance is defined as the response time perceived by the end user, i.e., the interval between theinstant at which an operator at a terminal enters a request for a response from a computer and the instantthe response is received at a terminal.
2Availability is the ratio of the total time a functional unit is capable of being used during a given intervalto the length of the interval.
Appendix 302
F.2. The IT Service support of Business Process
least 100 transactions a month just to pay for the security services company visit.
For ATM parts fault, the outsourcers are the hardware suppliers, which beside the pre-ventive maintenance, must also replace any malfunctioning ATM part. Four hours is ad-missible time between service desk notification and ATM recovery due to a malfunction.Each visit due to an emergency repair is charged at 100e. The regular maintenance costsof each ATM are 50e for each. All these costs must be taken in account when figuringglobal amount of operational costs.
F.2 The IT Service support of Business Process
The main IT support of the business process is the ATM system which communicateswith the bank’s central computer over an appropriate communication link. The ATM sys-tem has the following peripheral devices attached: a magnetic stripe reader for readingthe ATM card; a screen that displays messages to the user; a keypad that receives nu-meric input form the user; a deposit slot that receives deposit envelopes from the user; acash dispenser that dispenses cash (in multiples of 5e) to the user; a printer for printingcustomer receipts; and a key-operated switch to allow an operator to start or stop themachine.
Each ATM services one customer at a time. The customer is required to insert anATM card and enter a personal identification number (PIN) - both of which are sent tothe bank’s computer for validation as part of each transaction. To authenticate a userand perform transactions, the sent information is compared with the bank’s ac-countdatabase. For each bank account, the database stores an account number, a PIN and abalance indicating the amount of money in the account.
F.3 Main Activities of Business Process
F.3.1 Startup
The business process is ready for providing financial services to customers after the su-pervisor switches on the ATM system, enters the amount of money currently in the cashdispenser, and a connection is established with the bank. A message is sent by the ATMsystem to the service desk control system stating the beginning of operations.
The ATM maintains also an internal log of transactions to facilitate resolving ambigui-ties arising from a hardware failure in the middle of a transaction. Entries are made in thelog when the ATM is started up and shut down, for each message sent to the bank (alongwith the response back, if one is expected), for the dispensing of cash, and for the receiv-ing of a deposit or check. Log entries may contain card numbers and euro amounts, butfor security will never contain a PIN. A copy of each message is also send to the servicedesk control system that can provide actions from outsourcers regarding cash or suppliesreplenishment as well as replacement of malfunction parts.
Appendix 303
F.3. Main Activities of Business Process
Figure F.1: Business Process Model of Financial Services Provisioning
F.3.2 Open Financial Session
After the start up activity is finished, several financial sessions can happen. A financialsession begins when a customer inserts a magnetic card into the card reader slot of theATM machine that scans it. If the reader cannot read the card due to improper insertion ora damaged stripe, the card is ejected, an error screen is displayed, the session is aborted,the event is locally registered and a status code message is sent to the service desk controlsystem.
The customer is asked to enter his/her PIN, and is then allowed to perform a trans-action, by choosing one of the possible types. If the customer wants to make anotheroperation the menu is presented again allowing her/him to choose another transaction.If no more operations are desired, the customer can choose the exit option, to get the cardejected from the machine and the session ended.
The customer may abort the session by pressing the Cancel key when entering a PINor choosing a transaction type.
An invalid PIN is detected within a transaction when the bank reports that the cus-tomer’s transaction is disapproved due to an invalid PIN. The customer is required tore-enter the PIN and the original request is sent to the bank again. If the bank now ap-proves the transaction, or disapproves it for some other reason, the original request iscontinued; otherwise the process of re-entering the PIN is repeated. Once the PIN is suc-cessfully re-entered, it is used for both the current transaction and the next transactionsin the session. If the customer fails three times to enter the correct PIN, the card is per-manently retained, a screen is displayed informing the customer of this and suggestinghe/she contact the bank, and the entire customer session is aborted.
Appendix 304
F.3. Main Activities of Business Process
The customer will then be able to perform several transactions. The card will beretained in the machine until the set of transactions ends, at which point it will be re-turned to the customer - except as noted above for invalid PIN.
To summarize, the sequence of events are following:
1. The screen displays a welcome message and prompts the user to enter an accountnumber.
2. The screen prompts the user to enter the PIN.
3. The user enters a five-digit PIN using the keypad.
4. If the user enters the correct PIN for the account, the screen displays the main menu.If the user enters an incorrect PIN, the screen displays an appropriate message, andthen the ATM returns to Step 1 to restart the authentication process.
The ATM is able to provide the following services to the customer, displayed as op-tions of the main menu:
• A customer is able to make a cash withdrawal from any suitable account linked tothe card, in multiples of 5e. Approval must be obtained from the bank before cashis dispensed;• A customer is able to make a cash deposit to any account linked to the card. After
the amount of the deposit is entered into the ATM by the customer, the deposit issubject to currency recognition, acceptance, and recycling if it is cash. A manualverification by an operator is made in case of checks. The ATM counts the cash andgives a detailed receipt to confirm the deposited amount.• A customer is able to make a check deposit to any account linked to the card, in an
envelope. After the amount of the deposit is entered into the ATM by the customer,a manual verification by an operator is made in case of checks. The ATM counts thecash and gives a detailed receipt to confirm the deposited amount.• A customer is able to make a transfer of money between any two accounts.• A customer is able to pay routine bills, fees, and taxes (utilities, phone bills, social
security, legal fees, taxes, etc.), by entering data that identifies the document be-ing paid, such as the entity that issued the invoice, the payment reference and theamount to be paid.• A customer is able to purchase services (e.g. train, concert, movie tickets, adding
pre-paid cell phone / mobile phone credit, lottery tickets, etc.).• A customer is able to make a balance inquiry of any account linked to the card, and
print bank statements.• A customer is able to abort a transaction in progress by pressing the Cancel key
instead of responding to a request from the machine.• A customer is able to exit the system.
Appendix 305
F.3. Main Activities of Business Process
F.3.3 Withdraw Cash
A withdrawal operation is started within a session when the customer chooses this spe-cific transaction type from a menu of options.
The withdrawal transaction asks the customer to choose a type of account to with-draw from a menu of possible accounts, and to choose an amount in euros from a menuof possible amounts. The system verifies that it has sufficient money on hand to satisfythe request before sending the transaction to the bank. (If not, the customer is informedand asked to enter a different amount). The withdrawal transaction will then be sent tothe bank, along with information from the customer’s card and the PIN the customerentered. If the withdrawal is the operation the customer is doing and the bank reportsthat the customer’s PIN is invalid, the customer is required to re-enter the PIN and theoriginal request is sent to the bank again. If the bank now approves the transaction, ordisapproves it for some other reason, the original request is continued; otherwise theprocess of re-entering the PIN is repeated. Once the PIN is successfully re-entered, itis used for both the current transaction and the next transactions in the session. If thecustomer fails three times to enter the correct PIN, the card is permanently retained, ascreen is displayed informing the customer of this and suggesting he/she contact thebank, and the entire customer session is aborted, and the customer will not be offered theoption of doing another.
If the transaction is approved by the bank, the appropriate amount of cash is dis-pensed by the machine before it issues a receipt. (The dispensing of cash is also recordedin the ATM’s log and service desk control system). Then the customer will be askedwhether he/she wishes to do another transaction. If a withdrawal transaction is can-celed by the customer, or fails for any reason other than repeated entries of an invalidPIN, a screen will be displayed informing the customer of the reason for the failure of thetransaction, and then the customer will be offered the opportunity to do another transac-tion.
All messages to the bank and responses back are recorded in the ATM’s log and alsosend to the service desk control system.
The following steps describe in detail the actions that occur when the user enters theoption make a withdrawal:
1. The screen displays a menu containing standard withdrawal amounts: 20e (option1), 40e (option 2), 60e (option 3), 100e (option 4) and 200e (option 5). The menualso contains an option to allow the user to cancel the transaction (option 6).
2. The user inputs a menu selection using the keypad.
3. If the withdrawal amount chosen is greater than the user’s account balance, thescreen displays a message stating this and telling the user to choose a smalleramount. The ATM then returns to Step 1. If the withdrawal amount chosen is lessthan or equal to the user’s account balance (i.e., an acceptable amount), the ATM
Appendix 306
F.3. Main Activities of Business Process
proceeds to Step 4. If the user chooses to cancel the transaction (option 6), the ATMdisplays the main menu and waits for user input.
4. If the cash dispenser contains enough cash to satisfy the request, the ATM proceedsto Step 5. Otherwise, the screen displays a message indicating the problem andtelling the user to choose a smaller withdrawal amount. The ATM then re-turns toStep 1.
5. The ATM debits the withdrawal amount from the user’s account in the bank’sdatabase (i.e., subtracts the withdrawal amount from the user’s account balance).
6. The cash dispenser dispenses the desired amount of money to the user.
7. The screen displays a message reminding the user to take the money.
F.3.4 Deposit Cash
A deposit cash operation is started within a session when the customer chooses this spe-cific transaction type from a menu of options.
A deposit transaction asks the customer to choose a type of account to deposit themoney from a menu of possible accounts, and to type in a euro amount on the key-board.The transaction is initially sent to the bank to verify that the ATM can accept a depositfrom this customer to this account. If the transaction is approved, the machine acceptsthe cash from the customer before it issues a receipt. Once the cash has been received,a second message is sent to the bank, to confirm that the bank can credit the customer’saccount. (The receipt of an amount of cash is also recorded in the ATM’s log and servicedesk control system.)
A deposit transaction can be canceled by the customer pressing the Cancel key anytime prior to inserting the cash for deposit. The transaction is automatically can-celled ifthe customer fails to insert the bank notes within a period of 2 minutes after being askedto do so.
The following steps describe in more detail the actions that occur when the usermakes a deposit:
1. The screen prompts the user to enter a deposit amount or cancel the transaction.
2. The user inputs a deposit amount or cancels using the keypad.
3. If the user specifies a deposit amount, the ATM proceeds to Step 4. If the userchooses to cancel the transaction, the ATM displays the main menu and waits foruser input.
4. The screen displays a message telling the user to insert the bank notes into thedeposit slot.
Appendix 307
F.3. Main Activities of Business Process
5. If the deposit slot receives a deposit within two minutes, the ATM credits the de-posit amount to the user’s account in the bank’s database (i.e., adds the depositamount to the user’s account balance). [Note: The deposit will be rejected whethera false bank note is detected in the deposit or a discrepancy exists between thecash amount introduced by the customer and amount counted by the ATM system.When neither of these events occurs, the bank appropriately updates the user’s bal-ance stored in its database and the money is recycled and immediately available forwithdrawal.] If the deposit slot does not receive an amount to deposit within thistime period, the screen displays a message that the system has canceled the trans-action due to inactivity. The ATM then displays the main menu and waits for userinput.
6. In the case of a valid deposit, a second message will be sent to the bank indicatingthat the customer has deposited a certain amount of cash. (If the customer fails todeposit the correct amount within the timeout period, or presses cancel in-stead, nosecond message will be sent to the bank and the deposit will not be credited to thecustomer.)
7. The ATM provides the customer with a printed receipt for the successful trans-action, showing the date, time, machine location, type of transaction, deposit’samount, and account balance.
F.3.5 Deposit Check
A check deposit operation is started within a session when the customer chooses thisspecific transaction type from a menu of options.
A check deposit transaction asks the customer to choose a type of account to depositto from a menu of possible accounts, and to type in a euro amount on the key-board.The transaction is initially sent to the bank to verify that the ATM can accept a depositfrom this customer to this account. If the transaction is approved, the machine acceptsan envelope from the customer containing checks before it issues a receipt. Once theenvelope has been received, a second message is sent to the bank, to confirm that thebank can credit the customer’s account - contingent on manual verification of the depositenvelope contents by an operator later. (The receipt of an check envelope is also recordedin the ATM’s log and service desk control system.)
A check deposit transaction can be canceled by the customer pressing the Cancel keyany time prior to inserting the envelope containing the deposit. The transaction is au-tomatically canceled if the customer fails to insert the envelope containing the depositwithin two minutes after being asked to do so.
The following steps describe in more detail the actions that occur when the usermakes a deposit:
1. The screen prompts the user to enter a total check deposit amount or cancel the
Appendix 308
F.3. Main Activities of Business Process
transaction.
2. The user inputs a total check deposit amount or cancels using the keypad.
3. If the user specifies a check deposit amount, the ATM proceeds to Step 4. If the userchooses to cancel the transaction, the ATM displays the main menu and waits foruser input.
4. The screen displays a message telling the user to insert a check deposit envelopeinto the deposit slot.
5. If the deposit slot receives a deposit envelope within two minutes, the ATM creditsthe deposit amount to the user’s account in the bank’s database (i.e., adds the de-posit amount to the user’s account balance). [Note: This money is not immediatelyavailable for withdrawal by the customer. The bank first must physically verify theamount of cash in the deposit envelope, and any checks in the envelope must clear(i.e., money must be transferred from the check writer’s account to the check recip-ient’s account.) When either of these events occurs, the bank appropriately updatesthe user’s balance stored in its database. This occurs independently of the ATMsystem.] If the deposit slot does not receive a deposit envelope within this timeperiod, the screen displays a message that the system has canceled the transactiondue to inactivity. The ATM then displays the main menu and waits for user input.
6. In the case of a valid check deposit, a second message will be sent to the bank indi-cating that the customer has deposited a certain amount of checks. (If the customerfails to deposit within the timeout period, or presses cancel instead, no second mes-sage will be sent to the bank and the deposit will not be credited to the customer.)
7. The ATM provides the customer with a printed receipt for the successful transac-tion, showing the date, time, machine location, type of transaction, check’s amount,and account balance.
F.3.6 Transfer Amount
A transfer transaction asks the customer to enter the number of an account to transfer to,and to type in a euro amount on the keyboard. No further action is required once thetransaction is approved by the bank before printing the receipt. A transfer transactioncan be canceled by the customer pressing the Cancel key any time prior to entering aeuro amount.
F.3.7 Query Balance
A query transaction asks first the customer to confirm the operation. In case of a positiveanswer the ATM retrieve the balance from the bank’s database. No further action isrequired once the transaction is approved by the bank before printing the receipt. An
Appendix 309
F.3. Main Activities of Business Process
inquiry transaction can be canceled by the customer pressing the Cancel key any timeprior to confirm the account inquire.
F.3.8 Shutdown
The ATM machine can only be turned off when it is not servicing a customer. So, thesupervisor must make sure that no customer is using the ATM system, and then he/shecan turn the switch to the off position. The connection to the bank is then shut down.A status code message is sent to the service desk control system conveying the regularATM’s shut down.
The operator is now free to remove deposited checks, and make the replenishment ofcash, toner, and blank receipts, or do any maintenance of spares.
F.3.9 Spares Replacement
If the ATM’s device reader cannot read the card, the card is ejected, an error screen isdisplayed, the session is aborted, the event is locally registered and a status code messageis sent to the service desk control system.
This modeling example describes a business process implemented by financial institu-tions for providing financial self-services to customers, in public spaces. The businessprocess is a compound of human and automated activities, supported by a computer-ized telecommunications device known as automated teller machine (ATM).
With the implementation of this business process, the financial institutions have costsavings due to customers operating in self-service mode, i.e. without direct sup-port ofan institution’s employees, and also by benefiting from institution’s brand exhibition, aswell as the profits attained from advertising sold through ATMs.
To access the services, a customer must have an account, at one of the banks associatedto the ATM network, and use a magnetic card to interact with an ATM. Customers areable to do financial activities such as: to query their account balance, to withdraw cash(i.e., take money out of an account) and deposit funds (i.e., place cash or checks intoan account). Beside those automatic activities, human activities must also be done, inthe context of the business process to provide quality financial services to its customers.Among those activities we name a few such as cash replenishment, consumables (toner,paper) replacement, and deposit envelopes removal.
313
G.2. Business Process Overview
G.2 Business Process Overview
The business process is ready for providing financial services to customers after the su-pervisor switches on the ATM system, enters the amount of money currently in the cashdispenser, and a connection is established with the bank. A message is sent by the ATMsystem to the service desk control system stating the beginning of operations.
The ATM machine can only be turned off when it is not servicing a customer. So, thesupervisor must make sure that no customer is using the ATM system, and then he/shecan turn the switch to the off position. The connection to the bank is then shut down.A status code message is sent to the service desk control system conveying the regularATM’s shut down (see Figure G.1).
Figure G.1: Overview of the Business Process for Providing Financial Services
After the start up activity, a financial session begins after a customer inserts a mag-netic card into the card reader slot of the ATM machine that scans it. If the reader cannotread the card due to improper insertion or a damaged stripe, the card is ejected, an errorscreen is displayed, the session is aborted, the event is locally registered and a status codemessage is sent to the service desk control system (see Fig. G.2).
The customer is required to enter a personal identification number (PIN) - which issent, as well as the information in the card, to the bank’s computer for validation as partof the first transaction. The customer is asked to perform a transaction, by choosing one ofthe possible types. To authenticate a user and perform transactions, the sent informationis compared with the bank’s account database. For each bank account, the database storesan account number, a PIN and a balance indicating the amount of money in the account.The customer may abort the session by pressing the Cancel key when entering a PIN orchoosing a transaction type.
Appendix 314
G.2. Business Process Overview
Figure G.2: Detail of the Sub-Process Make Financial Operation
If the customer wants to make another operation the menu is presented again al-lowing her/him to choose another transaction. If no more operations are desired, thecustomer can choose the exit option, to get the card ejected from the machine and thesession ended.
In Figure G.3 are detailed the activities that are part of the Choose Operation sub-process. One of them is the sub-process Withdraw Cash, which is explained in more detailin next section.
Figure G.3: Detail of the Sub-Process Choose Operation
Appendix 315
G.2. Business Process Overview
G.2.1 Sub-Process Withdraw Cash
A withdrawal operation is started within a session when the customer chooses this spe-cific transaction type from a menu of options.
The customer is asked to choose a type of account to withdraw from a menu of pos-sible accounts, and to choose an amount in euros from a menu of possible amounts. Thesystem verifies that it has sufficient money on hand to satisfy the re-quest before send-ing the transaction to the bank. (If not, the customer is informed and asked to enter adifferent amount). The withdrawal transaction will then be sent to the bank, along withinformation from the customer’s card and the PIN the customer entered.
If the bank reports that the customer’s PIN is invalid, the customer is required to re-enter the PIN and the original request is sent to the bank again. If the bank now approvesthe transaction, or disapproves it for some other reason, the original request is continued;otherwise the process of re-entering the PIN is repeated. Once the PIN is successfully re-entered, it is used for both the current transaction and the next transactions in the session.If the customer fails three times to enter the correct PIN, the card is permanently retained,a screen is displayed informing the customer of this and suggesting he/she contact thebank, and the entire customer session is aborted, and the customer will not be offered theoption of doing another.
If the transaction is approved by the bank, the appropriate amount of cash is dis-pensed by the machine before it issues a receipt. (The dispensing of cash is also recordedin the ATM’s log and service desk control system). Then the customer will be askedwhether he/she wishes to do another transaction.
If a withdrawal transaction is canceled by the customer, or fails for any reason otherthan repeated entries of an invalid PIN, a screen will be displayed informing the customerof the reason for the failure of the transaction, and then the customer will be offered theopportunity to do another transaction.
All messages to the bank and responses back are recorded in the ATM’s log and alsosend to the service desk control system.
G.2.2 Required Modeling Work
The work required is that you make a BPMN model of the Withdraw Cash subprocessbased upon the information above, using the tools and the instructions provided in moredetail by the experience monitor.
G.2.3 Proposed Solution for the Modeling Case
Appendix 316
G.2. Business Process Overview
Figure G.4: Detail of the Sub-Process Withdraw Cash
Figure G.5: Detail of the Sub-Process Receive Withdraw Data
Figure G.6: Detail of the Sub-Process Handle Withdraw Approval
Appendix 317
G.2. Business Process Overview
Figure G.7: Detail of the Sub-Process Finalize Withdraw
Appendix 318
HA Process-Oriented Approach for
BPMN modeling
In a world driven by quality where organizations are moving to quality frameworkssuch as Total Quality Management (TQM) or Six Sigma, we need to assess modelingprocesses quality, as also done with data quality [WW96], and software process quality[ISO04a, SEI10c, SEI10b, SEI10a]. Quality in processes is a multidimensional concept in-herently perceived during process execution through processes’ outcomes1. For measur-ing process qualities, thresholds need to be specified, against which quality conformanceshould be evaluated. The process’ quality is the degree to which the quality characteristicsof the processes’ outcomes satisfies the stated and implied needs of its various stakehold-ers, and thus provide them the expected value.
The choice of the relevant process quality properties2 is primarily based upon thedomain where the process elicitation is done. For instance, in the realm of IT services[BPV05], some of the mentioned process execution quality characteristics (e.g. availabil-ity, responsiveness, security) [SMJ00] are specified and assessed through the Service-Level Management (SLM) process using suitable metrics [Smi08].
Having taking a product-oriented approach in the current dissertation, we intend asfuture work to pursuit a process-oriented approach to quality in BPMN modeling.
The attention focus of the process-oriented approach is the definition and assessmentof the set of quality attributes that processes’ outputs must attain. Therefore, alongside
1In this context, the reference to processes quality, means processes’ outcomes quality characteristics either atprocess design-time or execution-time.
2The qualities of a process are also known as constraints, quality attributes, quality characteristics [SG05],Non-Functional Requirements (NFRs), and softgoals [CNYM00].
319
information regarding the characteristics of outputs, the processes specification, accord-ing to this approach, should also contain information about the relevant domain depen-dent quality characteristics (e.g. response time, availability, accuracy) of the providedproducts and services. In line with the process-oriented approach, processes models,as outcome of process modeling, should also provide information regarding the valueattained by the stakeholders, i.e., the beneficiaries of the processes.
Current BPMN standard has a lopsided emphasis in the functional aspects (i.e. control-flow, data) of processes, without explicit mention both the goals to be achieved, and thequality characteristics of the provided outcomes, which is faced as technical issues left tobe handled at the level of the supporting systems.
To accomplish the aim of using BPMN for the specification and assessment of pro-cess quality characteristics, one must improve BPMN with additional meta-elements ad-dressing the process context. Hence, BPMN models should be complemented with theintentional information [Myl98] (e.g., goals, quality characteristics, metrics, measure-ment units, etc.), that enables processes’ quality specification at design phase, as wellas the subsequent assessment at run-time. The concept of goals is one of the conceptsfrom the intentional dimension, which defines the process context, by establishing whythe process is done and enabling to assess how good is the value delivered by the process[BPJ+10].
The verification of possible deficits of concepts in the BPMN regarding intentional di-mension can be done through an ontological comparison. However, the BWW, one of themost used ontologies for conceptual modeling analysis, does not cover the intentionaldimension. So, our first proposal is extending the BWW ontology to address the inten-tional dimension. The resulting extended BWW will be used to discover the constructsregarding processes’ quality not included in the BPMN metamodel. The derived eBWWontology will be used as framework for comparison and analysis of modeling languages(process modeling languages are a particular case), namely regarding the coverage by themodeling language concerning constructs required to represent quality characteristics.
Based on the ontological comparison, our second proposal will be a set of BPMNmetamodel extensions, with meta-classes and meta-associations. The meta-classes willsupplement the lack of constructs in BPMN that allow the specification and the followup assessment of softgoals, such as the elapsed time between request and outcome de-livery of a service (performance), the resources consumption for services delivered (ef-ficiency), or the period of time the service must be accessible to the requester (availabil-ity). Some other examples of relationships that we need to take into account are: (1) ifan activity has goals assigned, then the deliverables’ quality characteristics should bemeasurable and compared with the activity’s goals; (2) some quality characteristics mayimpact negatively with others (e.g. response time vs accuracy), so the trade-off amongthose characteristics must be established; (3) the correlation among resources and qualitycharacteristics should be specified, to know the amount of resources to be used to achievea certain level of a quality attribute.
Appendix 320
The drawback of the inclusion of more symbols covering aspects such as quality, ina language with rich semantics such BPMN, would be more complex process diagrams,to the point that would make the modeling language unusable. So, the challenge is howto address quality characteristics concerning processes without increasing the essentialcomplexity of BPMN models, thus jeopardizing its applicability.
Since process models can frequently change, we need a representation that ensuresthe abstractness and usefulness of the processes descriptions from domain specificationto systems implementation. This can be attained using a goals representation, whichallows capturing the reason for a process, at the design phase from a business perspective,as well as, the requirements for process execution at run-time phase. Therefore, the useof goals structures enable a top-down approach that starts on high-level business goalsand ends on operational goals, supported by low-level human or IT based activities.
To build up an intentional dimension on BPMN we should look for the researchdone till now, regarding goals, in the discipline of Requirements Engineering (RE), be-cause substantial work on RE emphasize the quality aspects besides functional require-ments of software artifacts [YM98, VL01, Kav02, KL04, RW05]. Of particular interest isthe perspective taken in Goal-Oriented Requirements Engineering (GORE), emphasiz-ing goals role in the system design process [MCY99]. In GORE, real-world problemsare largely non-functionally oriented (e.g., poor productivity, slow processing, high cost,low quality, and unhappy customer) [CL09]. GORE approaches with significant con-tributions include: the Goal-Based Requirements Analysis Method (GBRAM) [Ant96],the NFR Framework [MCN92, CN95, CNYM00, CL09], the Knowledge Acquisition inautOmated Specification (KAOS) methodology [DFL91, DvLF93, MHO06, vL09], the i*family, which includes: (1) the i* approach [YM94b, YM94a, YML96, Yu97]; (2) the Tropos[MKC01, GKMP04, BPG+04] [GMS05], and the extension Formal Tropos [FPMT01]; (3)the Goal Requirements Language (GRL) [UoT12, DHP05, MHO07].
By surveying GORE approaches, we expect to be able to assess the adaptation neededby methods from those approaches, to be used in the BPMN context [DSP09, DP11]. Wefound till now a main limitation in GORE approaches regarding how quantitative aspectsof requirements are dealt with: the measurement of different contributions of low-levelgoals and their weighting on the computation and measurement of top goals.
Hence, our third proposal is to derive a kind of graph based diagram, inspired byexisting state-of-the-art GORE approaches, instantiated from the extended BPMN meta-model. We want to convey with the graph diagram the network of goals to be achieved,the problems or obstacles to be avoided, the synergistic and conflicting goals, alternativesolutions to mitigate the problems, and the best alternative solution to be selected, as wellas the interrelationships of goals and quality characteristics.
An algorithm should be provided for computation of forecasted and actually reachedvalues of some particular quality attribute. The difference between the two values (actualand estimated) can be seen as a measure of the nonconformity or the degree of violationof agreed contracts with stakeholders. The label propagation procedure to be used in our
Appendix 321
approach, could be inspired although in a different way, from the one used in the NFRframework [CNYM00]. Our algorithm should be used both at the design phase for eval-uation of process specification and at run time, for assessment of possible violations ofpredefined quality characteristics based on collected data regarding the organization’soperations. In our proposal, relations between goals and activities should be established.A similarity between Tropos [MKC01] and our approach is that the goal structures aregraphs. An advantage of our proposal is that all used concepts (dynamic or intentional)are part of the BPMN language, so no other external language is needed for processes’goals specification and assessment.
For the sake of concreteness of our proposals we want to validate the results throughan empirical study regarding IT services. These kind of services are built upon the tech-nical infrastructure, as well as on systems and application software, to support processes.IT services can be depicted using BPMN models3. The specification and assessment ofgoals and NFRs for IT services, we are particularly interested in, are usually part ofService-Level Agreements (SLA) contracts [Sal04] among customers and IT providers.The goals and NFRs specify thresholds that should be accomplished by the providersregarding certain quality characteristics of IT services (e.g. maximum time to recover orrepair, response time, availability, accuracy, capacity and security). The SLA specificationand assessment of IT services, can be seen in the context of our proposals, as an instanceof the meta-problem of processes quality characteristics specification and assessment atdesign and execution phases of processes life cycle.
The specification and the assessment of processes’ quality characteristics will be vali-dated through a sample of five hundred thousand database records (BugZilla4) regardingtracking issues and bug reports of Mozilla Firefox browser downloaded from the MozillaFoundation web site5.
We have been publishing to date some work on processes’ quality, mainly focusedin its application to the realm of IT Services [FCBeA08, CBeA09, CBeA10a, CBeA10b,CBeAA11b, CBeAA11a].
3See for instance http://en.it-processmaps.com/products/itil-process-map-visio.html4http://www.bugzilla.org/5http://www.mozilla.org/foundation/
